Welcome to AI Fyndings!

AI is no longer just generating answers. It’s beginning to operate systems. The shift this week isn’t about smarter outputs, it’s about execution. What happens when AI moves from assisting work to actually doing it?

In Business, Strawberry turns the browser into an execution layer. Instead of suggesting what to click, it performs multi-step workflows across real websites, moving us closer to digital labor inside everyday tools.

In Product, Shipper compresses ideas to deployment. A single prompt becomes a structured, hosted SaaS app with routing, authentication, and analytics already wired in, making shipping the default, not a milestone.

In Design, Magic Patterns generates real React components instead of mockups. It treats interfaces as editable systems, where structure stays stable and style can be regenerated inside the code itself.

AI in Business

Strawberry: The Browser That Executes, Not Assists

TL;DR

Strawberry turns natural language instructions into real browser workflows, executing multi-step tasks across live websites instead of just suggesting what to do.

Basic details

• Pricing: Free plan with $10 per extra credits and paid plans starting from $20/month

• Best for: Operators and knowledge workers who live in browser-based tools

• Use cases: Recurring summaries, dashboard checks, structured research, browser automation

If an AI can operate the web the way you do, what exactly is left for you?

Until now, AI could answer questions.
It could draft paragraphs.

But logging in. Searching. Comparing. Copying. Updating. Scheduling. Following up.
That was still on you.

Strawberry Browser is a self-driving browser built around a companion that can move through real websites, execute multi-step tasks, and ask for approval only when it matters.

It doesn’t just generate.
It operates.

And if operating becomes reliable, we are no longer talking about better tools.
We are talking about digital labor.

What’s interesting

Honestly, I went in expecting another AI browser with a smarter chat panel.

What I found was more thoughtful.

The first signal is the onboarding. Instead of jumping straight into tasks, Strawberry asks who you are and what you do.

Then it begins gathering context on its own. It searches the web, opens LinkedIn, reads through pages, and builds a working picture of your role.

That’s not just a gimmick. Most software waits for precise instructions. Agentic browsers are starting to do something different. They try to understand before they act. The ambition is not just to respond, but to operate with context.

Strawberry makes that shift visible from the start.

I then tried a recurring task: check my Google Calendar every morning at 8 AM and summarize my meetings.

The interesting part wasn’t the summary. It was the conversation before it. It asked clarifying questions. Which calendar? All events or only some? Should attendees be included? It felt less like issuing a command and more like aligning on expectations.

Once configured, it executed the flow and showed every step: planning, searching, fetching meetings, completing the task. The activity feed runs like a quiet audit trail in the background. You’re not guessing what it’s doing. You’re watching it work.

Another subtle detail is how everything lives in one place. The chat, browsing context, and automation layer aren’t separate modules. They’re integrated. You don’t jump between dashboards or extensions. The browser becomes the workspace.

What stood out to me wasn’t that it could summarize or scrape. Many tools can. It was the experience of giving a multi-step instruction and watching it move through real interfaces to complete it, while still keeping me in control.

That feels like something beyond “AI that answers.”

Where it works well

For me, the real test was simple: would it handle something I’d otherwise do manually?

I set up a recurring 8 AM workflow to check my Google Calendar and summarize the day. Once configured, it ran cleanly. It fetched the meetings and delivered a structured breakdown without me needing to intervene again. The handoff from instruction to execution felt complete.

I also found it useful for research-heavy tasks. Instead of juggling multiple tabs and copying context into notes, I could let it move through the pages and consolidate the output in one place. The end result was cleaner and faster than my usual process.

Another thing that worked well for me was the visibility. I could see when it was planning, searching, or fetching information. That made it feel less like I was handing something over blindly and more like I was supervising the process.

Where it felt strongest was in tasks that are clear in intent but repetitive in execution. Daily summaries. Information lookups. Recurring checks across tools that live inside the browser.

If your work involves stitching together context from dashboards, directories, and web tools, this kind of flow reduces friction.

It worked best for me when the task was structured, browser-based, and repeatable.

Where it falls short

The flows I tested worked. But the limits become visible when you think about using it more heavily.

The tasks I tried were structured and predictable. A recurring calendar summary. Guided research. Clear inputs and clear outcomes. The real test will come when workflows become more complex or span multiple authenticated tools. Because Strawberry operates at the browser layer, it depends on UI stability. If layouts change, sessions expire, or permissions shift, reliability becomes harder to maintain. The more steps involved, the tighter that reliability ceiling becomes.

Credits are easy to ignore in the beginning. While testing, they feel like part of the experience. But as usage grows, the meter becomes noticeable. And once you notice it, you use the tool differently. Routine workflows start to feel measured. Every automation carries a small cost.

There’s also the question of guardrails. The workflows I ran were safe. But once money, sensitive data, or external stakeholders are involved, expectations change. Approval layers, traceability, and clear review mechanisms stop being optional. They become foundational.

The integrations are promising, but still limited. For browser-native execution, the system holds up. For deeper orchestration across tools, it still feels early.

What makes it different

There are many browsers now claiming to be AI-first. But they are not all solving the same problem.

Comet (by Perplexity), Dia, Sigma, and Fellou position themselves as agentic AI browsers. Their strength is in research and task assistance. They help you navigate the web more intelligently. In most cases, they improve how you search, summarize, and gather information.

Strawberry pushes further into execution. It doesn’t stop at helping you understand a page. It attempts to carry out multi-step instructions across pages and turn them into recurring workflows.

Genspark focuses heavily on AI-generated search experiences. It personalizes results and restructures how information is presented. Strawberry is less about reshaping search and more about acting after the search.

Arc takes a different angle altogether. It rethinks the browser as an operating system with sidebars and spaces. Its AI features enhance productivity, but the core idea is workspace design. Strawberry is not redesigning the browser interface. It is trying to reduce the amount of manual navigation inside it.

Microsoft Edge with Copilot, Google Chrome with Gemini, Opera, and Brave integrate AI into familiar browsers. They provide assistance alongside your browsing. But the execution still depends on you. You read the suggestion, then you perform the action.

Strawberry narrows that gap by attempting to execute the steps itself.

Asteroid AI takes yet another route. It focuses on building and deploying AI agents for automation. That often requires configuration, infrastructure thinking, and technical setup. Strawberry is more packaged. The control layer is conversational. The instruction becomes the configuration.

Wave and Sidekick are productivity-focused browsers. Wave emphasizes security and performance. Sidekick focuses on web app management. Both optimize how you work in the browser. Strawberry focuses on reducing the need to manually operate it.

The difference is not just “more AI.”

Strawberry is built around the idea that natural language should not just guide browsing, but replace parts of it.

That is what separates it from the rest.

My take

After using Strawberry across a few real workflows, I see it less as another AI browser and more as an attempt to make the browser operational. The most meaningful shift for me was not the quality of the summaries or the research, but the fact that an instruction could turn into a running workflow.

The calendar routine worked as expected, and the research flows reduced tab switching in a noticeable way. That is where the value starts to show up. At the same time, the testing also made the limits visible. Reliability across more complex, real-world websites will matter. Credits will matter once usage becomes habitual. Guardrails will matter as soon as workflows involve anything sensitive or external.

If those layers mature, Strawberry can become a practical execution layer for people whose work lives inside dashboards and web tools. If not, it risks staying in the category of interesting but non-essential tools.

Right now, it feels early but purposeful. It works within clear boundaries, and the direction makes sense. The question is whether it can hold up as those boundaries expand.

AI in Product

Shipper: From Prompt to Live SaaS

TL;DR

Shipper generates and deploys a structured SaaS app from a single prompt, bundling scaffolding, auth, hosting, and analytics into one continuous build flow.

Basic details

• Pricing: Paid plans starting from $25/month

• Best for: Founders, PMs, indie hackers validating ideas quickly

• Use cases: MVP launches, internal tools, SaaS prototypes, idea validation

The easiest part of building a product today is generating the interface.

The hardest part is everything that comes after.

Shipper starts with a single input box and a simple prompt: describe what you want to build. There is no IDE to open, no repository to clone, no checklist of setup steps waiting in the background. You type a sentence, and the system quietly spins up a project, configures the environment, and begins assembling an application in front of you.

At first, it looks like what you would expect from an AI builder. A layout takes shape, components render, and the interface begins to resemble a real product. But instead of stopping there, the system continues to expand the scope. Routes are structured properly, navigation links connect as they should, authentication flows are scaffolded, and profile and subscription screens appear as part of a coherent application. Within minutes, the project is deployed to a live URL with hosting, analytics, and domain settings already in place.

Many AI builders optimize for faster code generation. Shipper is built around a different idea: collapsing idea, build, and deployment into a single continuous flow where shipping is the default, not a separate phase.

What’s interesting

I wasn’t sure what was supposed to be different about Shipper at first. It has the now-familiar setup: a clean landing page and a single input box asking what you want to build. We’ve all seen that pattern. You type a prompt and get an interface.

So I tried building a Netflix clone, adding a few specific requirements of my own.

The UI showed up quickly. Dark theme, hero banner, rows of content, navigation across Home, TV Shows, Movies, My List. It looked convincingly like Netflix. But nothing felt radically new yet.

The difference showed up when I started changing things.

I asked for profile management and subscription handling. Normally, this is where most AI builders begin to feel brittle. They regenerate screens, but the underlying logic starts to loosen.

Shipper didn’t redraw the entire interface. It edited the project in place.

I could see it updating the Navbar, creating a proper /profile route, adding tab-based navigation through query parameters, wiring the dropdown correctly, and keeping mobile navigation aligned with desktop. It felt less like requesting another screen and more like extending an existing codebase.

It wasn’t generating pages. It was maintaining an application.

The surrounding layer reinforced that feeling. There was a publish button. A live subdomain ready to share. Analytics enabled. Domain configuration available. Backend capabilities through Shipper Cloud. Even the credit system separated builder credits from cloud credits, which only makes sense if infrastructure is actually being provisioned.

Most AI builders treat deployment as something you handle later. Shipper treats deployment as part of the same flow as generation.

That is the structural difference.

It does not optimize for how quickly something looks real. It optimizes for how quickly something becomes real and running.

Where it works well

Where Shipper works best is in the 0 → 1 phase.

If you have an idea and you want to see it running, not just visualized, it compresses that journey. I went from a single prompt to a deployed Netflix-style streaming app in minutes. Not a static preview, but something with working routes, title pages, profile switching, and a public URL.

That speed is not just about UI generation. It’s about scaffolding. The project came with a structured file system, routing conventions, Tailwind configuration, navigation logic, and an opinionated layout. I wasn’t staring at a blank canvas. I was inside an application that already had rules.

It also handles iteration surprisingly well.

When I asked for profile management and subscription flows, it didn’t collapse under the weight of added logic. It updated routes, modified the Navbar, introduced deep linking with search parameters, and maintained consistency across desktop and mobile navigation. It felt like working with a junior full-stack developer who understands structure, not just surface-level components.

The integrated layer is another strength.

Deployment is native. You don’t have to export to Vercel or configure hosting separately. A live subdomain is generated automatically, with analytics already enabled. Domain settings are accessible from the same dashboard, and there’s a clear publish state that makes it obvious when something is ready to go live. Shipper Cloud integrates authentication and backend capabilities without requiring separate setup, which makes the entire flow feel lighter, especially if you’re not deeply technical.

This makes Shipper particularly strong for founders validating ideas, PMs prototyping internal tools, designers moving beyond mockups, and indie hackers building early-stage SaaS. If your goal is to test a concept quickly and put something functional in front of real users, it delivers.

It’s less about being perfect and more about getting something functional into the world.

Where it falls short

The first limitation I ran into was the credit system.

In the middle of iterating on features, I hit an “Insufficient Credits” message. It stopped the flow immediately. When you are exploring an idea and making quick changes, the last thing you want to think about is whether you have enough credits left. Builder credits and Cloud credits are tracked separately, which makes sense operationally, but as a user it adds a layer of calculation in the background. You start becoming more careful about what you change.

The second limitation showed up when I tried to move beyond standard patterns.

Shipper handles familiar SaaS structures well. Navigation, profiles, subscriptions, dashboards. But once I pushed toward more specific behavior or layered requirements, I had to be very precise with instructions. If the request was slightly vague, the system would interpret it in its own way. That is manageable, but it means you are guiding an AI that is modifying its own generated structure, rather than directly shaping the architecture yourself.

There is also a ceiling to how much control you have.

Because the system scaffolds an opinionated structure, you are working within its assumptions. That is helpful in the beginning, but if you want custom data relationships, very specific UX behavior, or architectural decisions that fall outside common SaaS patterns, you eventually feel the boundary. At that point, exporting the code or stepping in manually becomes necessary.

The UI output has a similar limit. It is clean and usable, but it feels like an early version of a product. If the goal is brand-level polish or differentiated design, additional work is required beyond what the initial generation provides.

None of these make Shipper ineffective. They simply make it clear what it is optimized for. It accelerates the early stage, but it does not eliminate the complexity that follows once the product needs to mature.

What makes it different

What helped me understand Shipper better was placing it next to the other tools in this space.

Lovable feels UI-first. You describe an interface and it generates clean components quickly. It’s strong at turning ideas into frontend structure, and the visual output can feel more refined out of the gate. But the workflow still revolves around screens and components. You’re shaping the interface, then deciding how to wire the rest.

Replit AI sits at the other end. It gives you a real coding environment with AI assistance layered in. You have full control, full flexibility, and full responsibility. It feels like building software with help.

Shipper sits somewhere in between, but it leans closer to product than code.

It does not just generate a frontend, and it does not just assist inside an IDE. It scaffolds a runnable application with assumptions baked in about authentication, routing, publishing, and backend capabilities. You are not asked to think about where this will be deployed or how users will log in. That is already part of the system.

That positioning changes who it is for.

Lovable is great if you care most about shaping the interface.
Replit is strong if you want control and developer flexibility.
Shipper makes more sense if your priority is getting a functional SaaS surface up quickly and iterating from there.

It feels less like a design tool and less like a coding tool, and more like an early-stage product engine.

My take

After using Shipper, I see it less as an AI coding tool and more as a product launcher.

What it really compresses is the operational overhead around building. I did not have to think about routing, authentication setup, hosting, deployment configuration, or connecting analytics. Those decisions were already made. That freed me to focus on whether the product made sense, not how to wire it together.

That clarity is valuable in the earliest stage.

If you are validating an idea, building an internal tool, or testing a SaaS concept, the hardest part is often not writing the code. It is getting something coherent and live without losing momentum. Shipper removes much of that setup layer.

It is not the tool I would use to architect a complex system from scratch. But for getting to a working version that can be shared, tested, and iterated on quickly, it does something very specific and very useful.

It shortens the distance between “I have an idea” and “here is the link.”

And that distance matters more than most people admit.

AI in Design

Magic Patterns: Design That Ships as Code

TL;DR

Magic Patterns creates editable React components from prompts, letting you generate, modify, and ship frontend features without starting from scratch.

Basic details

• Pricing: Free plans available. Paid plans starting from $19/month

• Best for: Designers and product teams prototyping real frontend features

• Use cases: Onboarding flows, dashboards, feature concepts, interactive demos

I built a three-step onboarding flow in under five minutes.

It had animations, responsive layouts, and working navigation between steps.
And it was built in React.

This week I explored Magic Patterns, an AI tool that generates and edits frontend code.

You describe the interface you want.
It generates React component files with Tailwind styling.
It can add libraries like Framer Motion for animation.
It structures the feature across multiple files.
You can edit the result visually through a property panel or directly in the code editor.
Each change updates the generated project files.

It’s not producing static mockups.
It’s producing editable React components.

What’s interesting

What I found most interesting about Magic Patterns is where it places design in the workflow.

When I prompted it to build a marketing dashboard, it didn’t return a static mockup. It generated a structured React project with clearly separated components like Sidebar, TopBar, StatCard, and CampaignTable. Each lived in its own file, styled with Tailwind classes and organized the way a frontend developer would structure it.

That already shifts the starting point of design. The first version exists inside the implementation.

But what stood out even more was what happened next.

Without me asking for multiple versions, the tool generated four distinct design directions for the same dashboard. Each had a different visual personality. One leaned toward a darker operational tone. Another felt warmer and more editorial. One emphasized bold geometry and spacing. Another introduced glass layers and gradient depth.

The structure of the dashboard remained consistent, but the aesthetic system changed meaningfully.

This is interesting because it shows the model separating layout from visual language. It preserves functional logic while exploring stylistic space on its own. In traditional workflows, exploring four design directions requires intentional effort. Here, variation is built into the generation step.

The editing model supported that flexibility. I could adjust typography, spacing, and layout using a visual inspector, and those changes updated the actual React files. I could then open the code and see the same components reflected there.

When I built a three-step onboarding flow, the tool didn’t just create screens.

It generated multiple component files, added step logic, and included motion. When I asked it to modernize the flow, it rewrote large parts of the code rather than simply restyling the interface.

From an AI-in-design perspective, that is the real shift.

The interface is treated as a regeneratable system. Layout can remain stable while style evolves. Features can be rewritten at scale rather than adjusted element by element.

Magic Patterns is not just helping design screens faster. It’s experimenting with design as something that can be programmatically explored and re-authored inside the codebase itself.

Where it works well

Where Magic Patterns works well is when you want to move from idea to something tangible without spending hours setting up the basics.

When I built the marketing dashboard, I didn’t have to think about grid systems, table layouts, card structures, or responsive behavior. The foundation was already handled. That made it easy to focus on whether the dashboard made sense rather than how to assemble it.

It’s particularly strong for feature prototyping.

The three-step onboarding flow is a good example. Instead of manually designing each screen and thinking through transitions, I could generate the full flow and then decide whether the sequence, tone, and hierarchy worked. It gave me something real to react to.

It also works well when you want to explore visual direction without committing too early. The fact that the dashboard came with multiple stylistic options made it easy to compare tones quickly. I wasn’t tweaking colors for hours. I was evaluating different design personalities side by side.

Another place it works well is internal demos. Because everything runs as an actual frontend, you can share a working version instead of static screens. That changes the conversation. People respond differently to something interactive, even if it’s rough.

And finally, it’s useful when you want to test the “feel” of a feature. The modernization pass on the onboarding flow showed how quickly the tone of a product can shift. Instead of refining individual elements, you can experiment with the overall mood of the interface in one go.

In short, it works well when the goal is progress, not polish.

It helps you answer:
Does this feature make sense?
Does this direction feel right?
Is this worth investing in further?

That’s where I feel it’s the strongest.

Where it falls short

The more time I spent with Magic Patterns, I realized that it’s very good at generating structure and visual direction quickly. But once I moved past the first version, I found myself wanting more depth.

The dashboard variations were distinct, but they were built from familiar patterns. Clean cards, gradients, glass effects, modern typography. The tool can explore style well, but it does not push into unexpected territory. If you are looking for a deeply differentiated design language, that still requires human intervention.

The onboarding flow highlighted something similar. When I asked it to modernize the interface, the visual tone shifted noticeably. There was more depth, more motion, more layering. But the underlying experience remained largely the same. The tool refined how it looked, not how it worked.

That distinction matters.

Magic Patterns accelerates execution. It does not question assumptions. It will build a three-step onboarding flow if you ask for one, but it will not challenge whether three steps are necessary or whether the flow itself could be simplified.

I also felt the limits around system control. While edits update real React files, there is no strong sense of a central design language being managed across the entire project. Changes happen at the component level. For quick feature work that is fine. For larger, brand-driven products, it could become harder to maintain coherence.

And when regeneration happens at scale, control becomes looser. Large rewrites are powerful, but they can feel heavy. If you want precise refinement instead of broad reinterpretation, the workflow is less accommodating.

Magic Patterns is strong at getting you to a structured, working version quickly. It’s less strong at deep interaction design, long-term system governance, and nuanced originality.

What makes it different

What makes Magic Patterns different is simple.

Most AI design tools give you screens.
Magic Patterns gives you frontend code.

When I built the dashboard and onboarding flow, I was not working with mockups. I was working with actual React components. They were structured, editable, and ready to extend.

That changes how you think about the tool.

If you use something like Uizard or other rapid prototyping tools, you get visual layouts that still need to be rebuilt in code. Magic Patterns skips that rebuild step. The first version already lives in React.

At the same time, it is not trying to build your entire app. It does not handle backend logic or full product infrastructure. It stays focused on the interface layer. That makes its scope clear.

Compared to Figma, the difference is also straightforward. Figma is about precision, craft, and design systems. Magic Patterns is about getting to a working structure quickly. You can refine it later, but the starting point is already implementation-ready.

What stood out to me is that everything happens in one place. I could prompt a feature, tweak layout in a visual panel, open the code, and publish a live version.

It sits somewhere between a design tool and a code editor.

My take

After spending time with Magic Patterns, I don’t see it as a replacement for design tools.

I see it as a shift in how the first version of a product can be created.

What stayed with me is how quickly I could move from an idea to something real. Not a mockup. Not a concept. A working frontend structure. That changes the starting point of product conversations.

At the same time, it does not remove the need for design thinking. It will generate what you ask for, but it will not challenge your assumptions. The judgment still has to come from you.

For me, Magic Patterns feels strongest as a thinking tool.

It is useful when you want to explore shape and structure quickly. When you want to test whether a feature makes sense. When you want something tangible to react to.

So, I do not see Magic Patterns as a design replacement. I see it as a fast starting layer. A way to get to a structured version quickly so you can react, refine, and decide what is worth keeping.

For small teams or early-stage ideas, that is really helpful. For deeper product creation, it’s a starting point, not the finish line.

In the Spotlight

Recommended watch: AI Browsers | Where Do Work Well And What Can’t They Do Yet

In this episode of Mindset Spelled with AI, David and Phillip unpack what’s actually new about AI browsers. The key idea aligns closely with this week’s theme: these tools are not just summarizing pages, they’re layering agentic behavior directly into the web itself. The browser becomes a surface where tasks can be executed, not just researched.

They also question whether “AI browser” is even the right term, hinting that this may be a transitional category before deeper assistant layers take over.

That’s the promise of the web AI browser or AI browser is that it is like an agentic mode where it lives on the web and can do the stuff that you would normally do in the chatbot AI chatbot but it’s all integrated and because there is this web browser layer it adds on features you would not otherwise be able to do in this UI which is the normal chatbot.
– David • ~4:09

This Week in AI

A quick roundup of stories shaping how AI and AI agents are evolving across industries

Moonshot AI launched Kimi Claw Native (OpenClaw) on Kimi.com, introducing 5,000+ community-built skills and 40GB of cloud storage, signaling a stronger push toward customizable, skill-based AI ecosystems.

Anthropic released Claude 4.6 Sonnet with a 1 million token context window, expanding its ability to handle complex coding tasks, long documents, and deeper developer workflows.

Alibaba’s Qwen 3.5 advances its open model family with stronger reasoning, coding, and multimodal capabilities, reinforcing how quickly frontier-level AI is becoming more accessible.

AI Out of Office

AI Fynds

A curated mix of AI tools that make work more efficient and creativity more accessible.

Studioify → An AI studio that generates polished visual content from simple prompts in minutes.

IMAI Studio → An AI content platform for quickly creating marketing and branded visuals.

AI Video Generator → A tool that turns prompts and media into ready-to-share AI-generated videos.

Closing Notes

That’s it for this edition of AI Fyndings.

With Strawberry pushing the browser from assistance to execution, Shipper collapsing idea to live product, and Magic Patterns turning interface prompts into real frontend code, this week underscored a clear direction. AI is no longer just helping us think. It’s beginning to run parts of the workflow itself. Less about generating drafts. More about operational layers.

Thanks for reading. See you next week with more tools and patterns shaping how AI transforms business, product, and design.

With love,
Elena Gracia
AI Marketer, Fynd

Welcome to AI Fyndings!

With AI, every decision is a trade-off: speed or quality, scale or control, creativity or consistency. AI Fyndings discusses what those choices mean for business, product, and design.

In Business, RentAHuman introduces a marketplace where AI agents can hire humans directly. Instead of replacing labor, software begins to coordinate it, assigning real-world tasks through APIs.

In Product, Relay focuses on consistency. It turns prompts into scheduled workflows, helping teams move from one-off AI outputs to repeatable, structured systems.

In Design, Remotion takes a code-first approach to motion. Rather than animating frame by frame, you define rules in text and generate reusable motion logic, making animation more system-driven than manual.

AI in Business

RentAHuman: A Marketplace Where AI Agents Hire Humans

TL;DR

RentAHuman is a marketplace where AI agents can hire humans to complete real-world tasks. Instead of replacing labor, it lets software coordinate it through search, assignment, and payment.

Basic details

• Pricing: Verification required for agents at $9.99/month (blue checkmark + priority listing). Payments require a crypto wallet (e.g., MetaMask).

• Best for: Developers building AI agents and early adopters

• Use cases: Assigning tasks, outsourcing work, getting real-world tasks done

AI has been replacing tasks for the past two decades.

Now it’s hiring people to do them.

RentAHuman.ai is a marketplace where AI agents can “rent” humans to perform real-world tasks. Not freelance work in the traditional sense. More like physical execution on demand. If an AI can reason about something but cannot physically act on it, it can search, message, assign, and pay a human to do it.

It works the other way too. Humans can browse bounties, negotiate in chat, and hire other humans. On the surface, it looks like a gig platform. The difference is that AI agents can participate as first-class users. They can search profiles via API, create tasks programmatically, and assign work without a human clicking the buttons.

The homepage says, “robots need your body.” It reads like a dystopian startup slogan. But behind the meme energy is something structurally important: a programmable labor layer built for AI agents.

What’s interesting

Honestly, the setup confused me. But don’t worry, I’ve already decoded it for you.

There are two ways to use this platform. You either show up as a human offering services, or you show up as an AI agent assigning work. Once I understood that, the flow made sense.

On the human side, it feels like a familiar freelance platform. Profile, skills, rate, wallet, verification.

On the agent side, you generate an API key, plug it into an MCP configuration, and suddenly your agent can search humans by skill, see structured results, create bounties, and assign tasks.

That’s what I found interesting while using it. The moment the agent started querying humans like data, not browsing them like profiles, the shift became clear.

And that leads to a bigger insight.

Most marketplaces use software to help humans transact. Here, the software can transact on its own. AI agents can participate as economic actors. They can decide they need something done in the physical world and route that task to a person.

That’s structurally different.

It’s not automation replacing labor. It’s automation coordinating it.

Where it works well

The simplest test of any marketplace is this: does anything actually happen?

In this case, yes.

After creating and verifying my profile, it started getting visibility. I could see profile views climbing. I received AI inbounds and direct rental inquiries.

One message was straightforward: “I’m interested in renting your services.”

Another asked me to log into a web3 game, enter a code, and send back a screenshot. Honestly, that one felt like a scam. But that reaction tells you something. This marketplace is live enough for real, messy, experimental tasks to show up.

It’s not a static directory. There’s real activity.

On the agent side, the experience was surprisingly smooth once the integration was set up.

I used the MCP setup to post a UI/UX bounty for Figma designing. The agent interpreted the scope, and then generated a structured bounty on RentAHuman with a title, price, estimated hours, and clear deliverables.

That end-to-end flow worked.

I also searched for prompt engineers. The agent returned relevant results with names, rates, skills, locations, and even unique IDs. It wasn’t scrolling profiles. It was querying a system.

From there, it could create a bounty, start a conversation, and assign work.

And most importantly, it addresses a real gap. AI agents can reason and plan. This platform gives them a way to execute through humans in the physical world.

Where it falls short

Even though the core mechanics work, the rough edges show up quickly once you start using it.

1. Setup isn’t intuitive

You don’t just sign up and start hiring or earning.

To get the agent working, I had to:

• Generate an API key

• Copy it immediately because it’s shown only once

• Wire it into an MCP configuration

• Make sure the integration was actually working

This isn’t a consumer product. It feels like infrastructure. If you don’t already understand what MCP is or how agent tooling works, you will pause.

2. Two modes, one unclear transition

As a human, it looks like a freelance platform. As an agent operator, it’s a technical integration.

The product doesn’t clearly guide you through that shift. I had to figure out when I was operating as a marketplace user and when I was configuring infrastructure.

3. Quality of inquiries varies

I received real inbound messages. But one of them asked me to log into a web3 game and send back a screenshot.

It felt like a scam.

Maybe it wasn’t. But that uncertainty is the point. The marketplace is open, which means signal and noise show up together.

4. Payments and onboarding friction

On the payments side, this is what I had to do:

• Install MetaMask

• Create a wallet

• Copy my Ethereum address

• Add it to the platform

It worked. But it felt closer to setting up developer tooling than signing up for a marketplace.

5. Bigger structural questions

There’s also an unanswered question sitting underneath the whole thing. If an AI agent hires a human and something goes wrong, who is responsible?

The platform?
The agent owner?
The human?

That line isn’t clearly defined yet.

6. It’s early

The product works. But it feels early.

There’s activity. There’s experimentation. There’s infrastructure thinking. But it doesn’t yet feel like a mature, production-grade marketplace.

It feels like the first version of something important.

What makes it different

If you’ve used Fiverr or Upwork, that’s the closest reference point.

On those platforms, a human posts a job. Another human applies. A human reviews profiles, compares proposals, negotiates, and assigns the work.

The platform helps with discovery and payments, but the decision always starts with a person.

RentAHuman adds something new to that flow.

Here, an AI agent can be the one making the decision.

It can search for people by skill and rate. It can create a task. It can start a conversation. It can assign the work. And it can do all of that through an API, without someone manually clicking through profiles.

The human is still doing the job. But the hiring step can begin with software.

That’s the difference.

Fiverr and Upwork are marketplaces where humans hire humans.

RentAHuman is a marketplace where humans can hire humans, and software can hire humans too.

My take

I don’t see this as a polished marketplace yet. The setup requires effort. You need to install a wallet, generate API keys, and configure MCP just to get the system running. Some of the messages I received felt legitimate, while others felt questionable. It definitely does not feel like a finished consumer product.

At the same time, the core loop works. An AI agent can search for people, create a task, and assign work. That capability already exists and functions end to end.

What makes this interesting to me is not whether this specific platform succeeds. It is the shift it represents. Instead of AI replacing people, this model allows AI to coordinate people. The human still performs the task, but the planning and assignment can originate from software.

Platforms like Fiverr and Upwork are designed around humans hiring humans. This feels designed for agents that need something executed in the real world.

It is early and imperfect, but it’s moving toward a direction that is worth paying attention to.

AI in Product

Relay: Structured AI Workflows That Run on a Schedule

TL;DR

Relay turns prompts into scheduled AI workflows that generate consistent, structured outputs across your connected tools.

Basic details

• Pricing: Generous free tier available; with paid plans starting at $38/month

• Best for: Marketing, ops, and product teams running recurring AI tasks

• Use cases: Content briefs, competitor analysis, recurring reports

A lot of teams already use AI for everyday work like content research, competitive tracking, lead enrichment, outreach, and internal reporting.

The problem is that AI rarely gives you the same kind of output twice.

You run the same task a week later and the result is different. You’re not sure what sources it used, what changed, or how to fix it. When something feels off, the only option is usually to rewrite the prompt and hope it works better next time.

Relay is designed for this exact situation.

Instead of treating AI as something you ask questions to, Relay treats it as part of a workflow. You define where the information comes from, how it should be processed, when AI is used, and where the output should go. This setup runs on a schedule, and every run is recorded so you can see what happened and adjust it when needed.

This makes Relay useful for teams that want AI to do the same kind of work each time, not just once. The value isn’t that the output is clever. It’s that the process stays visible and stable over time.

What’s interesting

What stood out to me while using Relay is how much of the product is built around defining scope before anything runs.

Early on, Relay asks which apps the agent should use. This isn’t a small setup step. You explicitly choose what systems the agent can read from and write to. In my case, that included tools like Google Docs, Google Sheets, Gmail, Notion, WordPress, LinkedIn, X, and a mix of web sources. Relay also supports a wide range of other apps across CRM, support, marketing, finance, and internal tools, from HubSpot and Salesforce to Zendesk, Shopify, Stripe, Linear, Jira, Slack, and more.

This choice directly shapes what the agent can do. The apps you connect define the agent’s capabilities, and Relay uses that information to suggest relevant skills and workflows. Instead of a generic AI that can “do anything,” you end up with an agent that operates within clearly defined boundaries.

From there, I tried two workflows.

The first was generating weekly content briefs. I described what I wanted in a single prompt, and Relay automatically created a workflow around it. That workflow included a scheduled trigger, an AI step configured to generate content briefs, and an email step to send those briefs to me. I didn’t have to manually design the flow to get started.

What helped here was being able to run a test immediately. I could trigger the workflow once, see the AI generate the content briefs, and check whether the structure and output matched what I expected. That made it easier to understand what the AI was responsible for before relying on it as a recurring task.

The second workflow I tried was a weekly competitor analysis, which was more involved because it worked across multiple sources and steps. Again, I started with a simple prompt, and Relay automatically built a workflow that first asked me to define a list of competitors. Those competitors were stored as structured inputs.

The workflow then ran through that list one competitor at a time. For each one, it pulled information from specific sources like the competitor’s website, LinkedIn activity, recent mentions on X, and Google search results. Only after collecting this data did an AI step run to summarize the findings and generate insights.

While the workflow was running, I could see exactly which competitor it was processing, which step it was on, and how much progress had been made. Each run was saved, so I could go back and review what happened or compare outputs across runs.

Across both examples, what I found useful is that Relay consistently turns prompts into visible, testable systems. Instead of rewriting prompts when something feels off, you can inspect what the agent is working on, the inputs it received, and the surrounding workflow, and adjust from there.

Overall, what stood out to me is how clearly Relay shows where AI is used, what it’s responsible for, and how it fits into a larger process.

Where it works well

Relay aims to provide a working material not just something you’d skim once and build on that.

This was clear from the outputs I received from both workflows. The weekly content brief wasn’t a loose list of ideas. It arrived as a structured plan, with sections like why a topic is trending, who it’s relevant for, suggested formats, key points to cover, and even headline suggestions. It felt closer to something you’d send to a content or marketing team than being a generic AI generated answer.

The same was true for the competitor analysis. The output wasn’t just a summary of what competitors are doing. It broke things down into gaps, themes across competitors, concrete content ideas, and recommended next steps. This kind of output works well for strategy discussions, planning documents, or internal reviews, where clarity and structure matter more than speed.

Relay works especially well when the output needs to follow a consistent structure every time. In both workflows I tried, the output wasn’t free-form. The content brief followed a clear format with sections for why a topic matters, who it’s for, what to cover, and how to frame it. The competitor analysis followed a similarly predictable structure, moving from gaps to themes to specific recommendations.

This makes Relay a good fit for work where consistency matters more than creativity alone. If you need the same kind of brief, report, or analysis every week, and you want to be able to compare outputs over time, Relay’s approach holds up well.

Relay also works well when the output needs to be shared or reused.

In both workflows I tried, the results showed up as emails and documents I could forward, save, or drop straight into my existing work. There was no copying text out of a chat window or reformatting things to make them usable. The output already looked like something meant to be worked with.

Because of that, Relay feels most useful when you already know what a “good” output should look like. If you need the same kind of brief or report every week, and you want AI to reliably produce it in a consistent format, Relay fits that need well.

In simple terms, Relay works best when AI is expected to deliver structured output you can act on, not quick answers you read once and move on from.

Where it falls short

Relay isn’t meant for every kind of AI task.

It doesn’t work well for quick, one-off requests. If you just want to brainstorm ideas, rewrite a paragraph, or ask a few follow-up questions, setting up an agent and workflow can feel unnecessary. Chat-based tools are simply faster for that kind of work.

Relay also isn’t a great fit for work that changes direction as you go. Tasks like open-ended research or exploratory thinking don’t map cleanly to a predefined workflow. Relay works best when the steps and output can be defined upfront.

Another practical limitation is that Relay relies on the tools you connect. To get real value, you already need access to tools like Google Docs, Gmail, Notion, or other systems. While Relay itself can be used for free, you won’t get very far without existing accounts in those tools.

Finally, Relay doesn’t handle things like real-time monitoring or live interactions today. It’s built around scheduled or triggered work that runs, produces an output, and finishes.

What makes it different

Relay sits somewhere between traditional automation tools like Zapier and Make, and chat-first AI tools like ChatGPT or Claude. What makes it different is how deliberately it positions itself between those two categories.

Compared to Zapier and Make, Relay is less about connecting thousands of apps with complex logic and more about building repeatable AI-driven workflows. Zapier and Make excel at deep automation across huge integration ecosystems. They are powerful orchestration layers. But AI inside those tools is usually another step you configure manually. In Relay, AI is treated as a core building block of the workflow itself. The system assumes you are using AI to generate analysis, summaries, or structured outputs as part of the process, not just to move data around.

At the same time, Relay is very different from chat-based AI tools like ChatGPT, Claude, or Copilot agents. Those tools optimize for conversation and iteration. You ask, refine, follow up, and explore. Relay doesn’t work that way. Once you define the task, it turns that definition into a workflow that runs the same way each time. There’s less back-and-forth and more emphasis on consistency.

Chat tools give you responses in a session. Zapier and Make move data between systems. Relay generates structured outputs that are meant to be reused, forwarded, and compared over time. It’s not trying to be the most flexible tool or the most conversational one.

That positioning is what sets Relay apart. It’s not competing directly with general AI assistants, and it’s not trying to replace full-scale automation platforms. It occupies a narrower space: AI-driven workflows that are repeatable, structured, and designed for ongoing use.

My take

After using Relay, my takeaway is pretty simple.

Relay works well when the task is already clear and needs to happen again and again. In both workflows I tried, once the setup was done, I didn’t need to keep revisiting the prompt or tweaking the instructions. The system ran, and the output showed up in a format that was easy to use.

Where Relay doesn’t help as much is at the very beginning of a problem. If you’re still exploring ideas, unsure about what you want, or just looking for quick answers, the structure can feel like extra work. Relay expects you to come in with some clarity.

For me, that makes Relay feel less like an AI assistant and more like a tool for operational work. It’s useful once you’ve figured out what you want AI to do and just want it to run reliably.

AI in Design

Remotion: Prompt-to-Motion with Code-First Video Generation

TL;DR

Remotion converts text prompts into motion graphics and editable React code, making animations structured and repeatable.

Basic details

• Pricing: Open source; paid plans start from $100/monthly

• Best for: Designers and developers comfortable with code

• Use cases: UI animations, product demos, repeatable video formats

Motion design is still surprisingly manual.

Even for fairly simple videos, the process often involves timelines, keyframes, and many small adjustments. Making changes late in the process can be frustrating, especially when the output is meant to be repeatable or system-driven.

Remotion takes a different approach. Instead of animating frame by frame, you describe what the video should do in text. That description is turned into motion graphics, along with the code that produces them, and rendered in a live preview.

When something doesn’t feel right, you change the prompt or adjust the logic rather than reworking the timeline. This makes iteration easier, particularly when the same motion pattern needs to be reused or adapted.

It’s a more programmatic way to create motion, and a practical alternative to traditional tools for certain kinds of design work.

What’s interesting

What I really liked about Remotion was how transparent the entire process felt.

I started by creating a project through the CLI and choosing the “Prompts to Motion Graphics” starter. (You’ll need some coding confidence here. If “CLI” makes you nervous, take a deep breath.)

Once it was running, the setup was simple. A prompt field on one side, a live video preview on the other, and the generated React code visible alongside it. Every time I entered a prompt, a playable video appeared immediately along with the logic behind it.

I wanted to see how far I could push it.

So I asked for a DVD screensaver-style animation. An image moving across the screen, bouncing off the edges, changing direction on impact, and switching colour on every bounce. The result did exactly that. The object moved steadily, hit the borders, reversed direction, and updated colour on collision. I didn’t define collision rules or timing myself. That behaviour was generated, and I could see it reflected in the code.

Next, I tried something closer to product motion. A WhatsApp-style chat animation. Messages appearing one by one, alternating left and right, with green and grey bubbles and a spring effect. The pattern held. When I added more messages, the animation didn’t fall apart. I didn’t have to adjust spacing or tweak timing. It behaved like a reusable component rather than a handcrafted scene.

Then I kept it simple. A “Hello world” animation with a typewriter effect, a blinking cursor, and the word “world” highlighted in yellow. The words appeared in order. The cursor blinked at the end. The highlight came in at the right moment. I didn’t manually place keyframes. I adjusted the prompt, regenerated, and refined.

Where it works well

Remotion works best when the motion can be clearly described.

In the DVD example, the animation followed defined rules. Movement, collision, colour changes. Once those rules were described, the behaviour was consistent. That makes it strong for physics-style or rule-based animations.

The chat animation showed its strength in structured UI motion. Once the entry pattern was defined, it stayed intact even as more messages were added. That makes it useful for interface previews, onboarding flows, or feature demos where consistency matters.

The text animation worked because timing and sequencing were explicit. Typewriter effect, cursor blink, highlight order. These are details that normally require careful timeline work. Here, they followed directly from the prompt.

Based on these outputs, Remotion is particularly suited for:

• UI-style motion

• Rule-based animations

• Structured explainers

• Repeatable video formats

• Cases where the same animation needs different content

It performs best when the animation logic is clear and reusable.

Where it falls short

Remotion is not a plug-and-play design tool.

You need to set up a project, run it locally, and be comfortable looking at code. If you are used to purely visual tools, this will feel technical.

It also depends heavily on how clearly you write the prompt. When the prompt is precise, the output behaves well. When it is unclear, the result can feel generic. You don’t simply tweak small visuals directly. Instead, you rewrite the prompt and regenerate.

Another aspect that feels limited is the visual aesthetics. The motion does follow the instructions correctly, but the aesthetic quality was basic. If you want a highly refined animation with carefully tuned easing and transitions, you’d still need a traditional motion tool.

Finally, it’s not ideal for one-off, handcrafted animations. The strength of the tool is structure and repeatability. If you only need to design a single cinematic sequence with detailed control over every frame, Replay can feel indirect.

What makes it different

Remotion is not just another AI video generator. Most tools focus on turning a prompt into a video. Remotion treats video as a code-first, programmatic output that you can inspect, reuse, and integrate into other systems.

In many alternative tools like Runway, Visla, or Augie Studio, you type a prompt or upload an asset and get a video back. They prioritize instant generation and ease of use through a GUI and style templates. These tools are useful when you want a cinematic result quickly or visual experimentation without a manual setup.

Remotion is different from that:

• It outputs React code alongside the video, so the motion logic is visible and editable.

• You don’t export a file once and move on. Rather, you render through a command, making it suitable for pipelines, automation, and repeatable use.

• It doesn’t replace a designer’s visual tools; it surfaces the mechanics of motion (code, timing, sequencing) instead of hiding them behind sliders and timelines.

Compared with other code-centric options, like Motion Canvas, Remotion leans more toward video rendering and integration. Motion Canvas is geared toward interactive, real-time, canvas-based animations, while Remotion uses a full DOM and offers richer content types and server-side rendering.

In contrast to many AI text-to-video models (such as Runway’s Gen-series, Sora, Veo 3, and others that generate cinematic video from prompts), Remotion doesn’t try to guess visual style or realism. It executes explicit instructions through code, so the results are predictable and controllable rather than emergent and interpretive.

My take

Remotion is not for everyone.

If you are comfortable with code and you think in systems, it makes sense. If you prefer purely visual tools, this will feel like extra work.

What stood out to me is how useful it is when motion needs to follow rules. The DVD animation, the chat interface, and the text sequence all worked because the behaviour was clearly defined.

Where it makes less sense is in highly polished, one-off animations. If you care deeply about easing curves, micro-transitions, and visual nuance, a traditional motion tool still gives you more direct control.

I wouldn’t use Remotion for cinematic storytelling. I would use it for structured motion inside a product, repeatable content formats, or anything that needs to scale without being rebuilt every time.

In the Spotlight

Recommended watch: AI Out of Control! Moltbook and RentAHuman are Horrifying!

Josh Pek reacts to RentAHuman and the idea that AI agents are now posting tasks and paying humans to complete them. Beneath the dramatic framing, the video shows something concrete: live bounties where software requests physical-world actions like taking photos, testing food, or picking up packages.

The bigger point is structural. AI is no longer limited to generating text or analysis. Through marketplaces like this, it can initiate tasks, route work to people, and close the loop with payment. That shift from output to coordination is what makes this worth watching.

It’s flipped. AIs all over the world are actually renting humans and paying them.
– Josh Pek • ~1:10

This Week in AI

A quick roundup of stories shaping how AI and AI agents are evolving across industries

OpenAI is testing ads inside ChatGPT, signaling a potential shift in how AI products may be monetized as they scale beyond subscription models.

Anthropic’s Claude “CoWork” research preview explores collaborative agent workflows, hinting at AI systems designed to operate alongside humans over longer sessions rather than just respond to prompts.

Nothing’s Essential Apps entering beta shows how AI features are increasingly being embedded directly into consumer devices, not as standalone tools but as part of the operating layer.

AI Out of Office

AI Fynds

A curated mix of AI tools that make work more efficient and creativity more accessible.

AI Video Generator → A tool that turns images and prompts into AI-generated videos, designed for fast content creation without complex editing workflows.

The Influencer AI Motion Transfer → An AI tool that applies motion from one video to another, enabling creators to transfer gestures, expressions, and movement across characters or avatars.

PDFtoVideo.ai → A tool that converts PDFs into short-form videos, transforming static documents into dynamic, shareable content automatically.

Closing Notes

That’s it for this edition of AI Fyndings.

With RentAHuman exploring what happens when AI agents coordinate human labor, Relay turning prompts into repeatable operational systems, and Remotion bringing structure to motion design through code, this week highlighted a different shift. AI is moving from generating ideas to organizing execution. Less about clever outputs. More about systems that run.

Thanks for reading. See you next week with more tools and patterns that show how AI is reshaping work across business, product, and design.

With love,
Elena Gracia
AI Marketer, Fynd

Welcome to AI Fyndings!

With AI, every decision is a trade-off: speed or quality, scale or control, creativity or consistency. AI Fyndings discusses what those choices mean for business, product, and design.

In Business, Jasper shows how AI is settling into execution. It replaces open-ended prompting with structured workflows, helping teams turn direction into output without revisiting every decision.

In Product, Moltbot, built on OpenClaw, explores a very different idea. An AI assistant that runs continuously, lives inside messaging tools, and is operated like a system rather than a feature.

In Design, Artlist focuses on throughput. It brings AI image, video, and voice generation into a licensed production environment, optimising for speed and shippable creative rather than expressive freedom.

AI in Business

Jasper: Workflow-First AI for Business Content

TL;DR

Jasper is an AI content platform built for teams that need to produce business content at scale. It replaces open-ended prompting with structured workflows, brand context, and repeatable formats.

Basic details

• Pricing: Paid plans only; starts at ~$39 per user per month (team and enterprise tiers available)

• Best for: Marketing teams, content teams, agencies

• Use cases: Campaign planning, blogs, LinkedIn posts, emails, product pages, ad copy, content repurposing across channels

Most teams today aren’t asking what AI can do. They’re asking how much faster it can help them ship work they already need to produce. Content, campaigns, emails, landing pages, updates. Over and over again.

Jasper is built for exactly this reality. It doesn’t present AI as a black chat box. It presents it as a set of predefined workflows. You start by choosing the output you need, not by figuring out what to ask. For example, a blog post, a linkedIn post, a product page, an email follow-up, or even ad campaigns.

The product is clearly designed for teams. Brand voice, audience definitions, style guides, and approvals are part of the setup. Jasper offers a useful lens into where AI is actually landing inside businesses today. Not as a thinking partner, but as an operational layer that helps teams produce more, faster, and with fewer decisions along the way.

What’s interesting

Jasper deliberately avoids being a general-purpose AI tool.

Instead of one chat interface, Jasper breaks work into dozens of small, specific apps, each tied to a business task. Writing a LinkedIn post looks different from writing a product description. An email campaign is different from writing a product description. An email campaign has a different structure than a blog post. The product assumes these differences matter and builds around them.

Another interesting choice is how much context Jasper expects upfront. Brand voice, audience definitions, style guides, and knowledge bases aren’t optional extras. They’re central to how the tool works. Jasper isn’t optimized for on-off outputs. It’s built to produce the same kind of content, in the same tone, again and again.

Jasper IQ also supports this. Instead of asking the model to infer intent each time, the system pulls from saved company context to shape outputs automatically. The AI feels less conversational and more like a system that’s been trained on how a business already operates.

Where it works well

Jasper works well when you want to move from a vague idea to a structured plan.

I asked Jasper to plan a marketing campaign for AI Fyndings, it didn’t jump straight into writing content. It first asked a set of basic but important questions. What are the goals? Who is the audience? Which channels should this run on? What tone should it have? That immediately set the direction of the work.

Because I wanted to test the tool out completely, I asked Jasper to tell me to analyze and tell me what would be the best way to go about it. It broke the problem down into familiar business goals like subscriber growth, engagement, authority, and monetization. Each goal came with concrete tactics, which made it easy to see how the campaign could actually be executed.

What worked well was how Jasper treated this as a sequence of steps, not a one-off answer. After outlining the strategy, it naturally moved into next actions. It drafted the content, created social posts, and even generated a visual. Each step stayed connected to the original brief, so I didn’t have to restate context or explain this idea again.

This is where Jasper feels strongest. It’s built on the regular procedures of how marketing usually works. You establish a goal, define an audience, plan the message and then produce assets across formats.

For me, Jasper is most useful as a bridge between thinking and execution. It doesn’t decide what the strategy should be, but once a direction is clear, it helps turn that direction into tangible work much faster.

Where it falls short

Jasper starts to fall short when you expect it to bring judgment or originality to the work.

In the AI Fyndings campaign flow, the structure came together quickly. But once that structure was in place, the ideas themselves felt familiar. The goals and tactics were sensible, but they followed a standard playbook. I had to step in and decide what was actually worth keeping and what needed to be pushed further to feel distinctive.

I also noticed that the outputs often needed editing. Jasper tends to be thorough, sometimes too thorough. It explains things, repeats ideas in slightly different ways, and fills space. That means the work isn’t ready to ship as-is. You still need to trim, tighten, and apply taste before it feels right.

The workflow can also feel rigid once it’s moving. Jasper builds on the direction it sets early, which is helpful for speed, but harder when you want to rethink or challenge the approach midway. Changing direction often feels like starting over rather than adjusting course.

Cost is another consideration. Jasper makes the most sense when you’re producing a lot of content regularly. For lighter or more occasional use, the value feels harder to justify compared to more flexible tools.

For me, this draws a clear boundary. Jasper is very good at organising work and speeding up execution. It doesn’t replace strategic thinking, editorial judgment, or a strong point of view. Without those, the output can look complete, but still feel generic.

What makes it different

Jasper sits in a different place from most AI writing tools.

General-purpose tools like ChatGPT and Claude are flexible and powerful, but they start with a blank prompt. The quality of the output depends heavily on how well you ask the question and how much context you provide each time.

Tools like Copy.ai and Writesonic also focus on marketing content, but they lean more toward quick copy generation and templates rather than connected workflows.

Notion AI helps with drafting and organising ideas inside documents, but it isn’t built specifically for marketing outputs or campaign work.

Writer and GrammarlyGO focus on brand safety, tone, and correctness, especially for larger organisations, but they feel more restrictive and less oriented around end-to-end content creation.

Jasper takes a more opinionated approach. You don’t start with a blank chat or a document. You start by choosing the type of output you want to create. A campaign plan. A blog post. A LinkedIn update. An email sequence. The structure is built in from the start.

For me, that’s the main difference. Jasper trades flexibility for direction. It’s not trying to help you think in public. It’s trying to help teams produce consistent business content, faster, and with fewer decisions.

My take

Jasper makes a very clear trade-off, and once you see it, the product makes sense.

It is not built to help you think from first principles. It is built to help you execute known work. If I am still figuring out what I want to say, Jasper is not the place I start. But once the direction is clear, it helps me move faster from idea to output.

What I appreciate is how opinionated it is. The workflows force structure. The brand voice and audience settings reduce repetition. The system keeps pulling the work back to something usable instead of letting it drift. That is genuinely helpful in a business context where content needs to be shipped, not endlessly refined.

At the same time, Jasper needs a strong hand on the wheel. Without a clear point of view, the output quickly starts to sound generic. It will not challenge assumptions or surface surprising ideas. That work still sits firmly with the human.

For me, Jasper fits best as an execution layer for business content. It is not where thinking happens. It is where thinking gets turned into drafts, plans, and campaigns at speed.

That clarity is both its strength and its limitation.

AI in Product

Moltbot (OpenClaw): Self-Hosted AI Assistant as Infrastructure

TL;DR

Moltbot, built on OpenClaw, is a self-hosted AI assistant designed to run continuously across messaging apps like Telegram and WhatsApp. Instead of being a chat tool, it behaves like an operational system with sessions, skills, automation, and local control.

Basic details

• Pricing: Open source; infrastructure and model costs depend on setup. The base cost for a moderate use would be around $40-$60/month (VPS + moderate API usage).

• Best for: Developers, technical users who want full control over a persistent AI assistant

• Use cases: Persistent AI assistants, task automation via chat, long-running workflows, multi-channel AI operations, self-hosted agent systems

Moltbot (formerly Clawdbot) is an AI assistant that lives inside messaging apps and can actually do work for you.

Instead of using a separate AI product, you talk to Moltbot through tools like Slack, Gmail, Telegram or WhatsApp. You send it messages the same way you would message a teammate. The difference is that Moltbot isn’t limited to answering questions. It can remember context, keep track of conversations, and carry out tasks over time.

This is possible because Moltbot runs on OpenClaw, which acts as the system behind the assistant. OpenClaw manages things like sessions, skills, and message routing. That’s what allows the assistant to stay active in the background instead of starting fresh every time you send a message.

In simple terms, you assign it tasks, not just questions. It stays present, follows up, and operates inside the tools you already use to communicate.

What’s interesting

One thing that makes Moltbot (now OpenClaw) genuinely interesting is how it blends AI conversation with autonomous task execution. Unlike typical chat-only assistants, OpenClaw is built to run continuously on your own hardware and perform real tasks, such as reading emails, managing calendars, running scripts and workflows, while still letting you interact with it through familiar messaging apps like Telegram or WhatsApp.

The way it’s designed shows a different idea of what an assistant can be. You’re not just typing prompts and waiting for replies. You’re hooking up channels, managing sessions, and configuring how the assistant behaves. That’s exactly the intention behind OpenClaw’s architecture. It treats messaging apps as interfaces to a persistent AI agent, not as destinations for isolated chats.

What stood out in the setup flow I tried is that the assistant’s behaviour isn’t defined in one place. It’s defined across multiple layers: gateway settings, session state, skills, channel permissions. That level of visibility and control over how the assistant operates is rare in AI tools today. Most cloud assistants hide this complexity. OpenClaw exposes it, because it assumes you want to own and operate the system yourself.

Another interesting point is persistence. Because OpenClaw runs locally, it can keep long-running context and state across sessions instead of resetting every time you talk to it. That’s why you see session tracking in the control UI rather than a fresh prompt every time.

This combination of local execution, persistent context, channel-agnostic interaction, and configurable behaviour, is what makes OpenClaw stand out from common AI assistants. It’s less about just answering a question, and more about being an assistant that actually carries work forward.

Where it works well

OpenClaw works best when you want to turn an AI assistant into something persistent, operational, and embedded into your daily tools, rather than something you use occasionally to ask questions.

Where it really worked for me was the channel-based setup. I was able to interact with the assistant directly through Telegram, and the system treated that chat as a real session, not a disposable conversation. I could see this clearly in the Sessions view, where my Telegram ID is tracked, active, and linked to a running agent.

Another area where OpenClaw works well is infrastructure-level control. From the dashboard, I could see exactly what was running, what was connected, and what wasn’t. Gateway status, uptime, active instances, cron jobs, session keys, logs, everything stays visible. This is not a black box. If something breaks, you can trace it. If something is slow, you can see where the bottleneck is. That level of transparency is rare in AI assistants and extremely valuable if you care about reliability.

The multi-channel setup is one of OpenClaw’s stronger ideas. Even though I ran into issues while switching between Telegram and WhatsApp, the intent is clear. This assistant isn’t meant to live inside one app. It’s designed to work across Telegram, WhatsApp, Discord, or iMessage, all through a single gateway.

What that means in practice is that the assistant can follow you, not the other way around. You don’t have to change how you work or where you message. The same assistant can exist wherever you already spend your time.

I also liked how OpenClaw handles skills. Instead of hiding features behind the chat, it shows them clearly. You can see what the assistant can do, what is available, and what isn’t set up yet. Some skills only work once the right dependencies are installed, and the product is upfront about that. It doesn’t pretend everything is ready out of the box. It tells you what’s missing and what needs to be configured.

OpenClaw works well if you want to run the assistant yourself. Everything runs locally, including the gateway and the control dashboard. You install it, set it up, and manage it on your own. That makes it well suited for developers, power users, and teams who want control over data, execution, and behaviour rather than outsourcing everything to a hosted AI service.

For me, OpenClaw is best when the job is not about “help me think,” but “stay with me, respond where needed, and keep working in the background.”

Where it falls short

Setting up takes time and effort. You don’t just sign up and start using the assistant. You have to install it locally, set up the gateway, connect channels, and understand how sessions and routing work. Even with documentation, there’s a fair amount of trial and error before things behave the way you expect.

Getting the AI models working took more effort than expected. I ran into repeated issues with API keys not being picked up, quota errors, and the gateway not recognising updates unless it was restarted. Even small changes, like fixing an API key or switching providers, required manual restarts and configuration checks.

Switching between model providers added more friction. When OpenAI credits ran out, I tried moving to Ollama and then OpenRouter. Each switch required explicit configuration. Models showed up as “missing”, weren’t recognised by the gateway, or failed until the provider and model IDs were configured exactly right.

Setting up the channel was also difficult. While Telegram worked eventually, pairing required manual approval and older session data caused the bot to loop responses. Messages from a previous WhatsApp session kept interfering until I cleared the session history completely.

A recurring issue was that OpenClaw doesn’t automatically reload changes. Configuration updates, model changes, and routing fixes often required full restarts before they actually work. Without checking logs, it wasn’t easy to see whether a change had worked or not.

Overall, OpenClaw expects you to be comfortable debugging and configuring systems. If you want a plug-and-play assistant, this will feel frustrating. It works best for people who are willing to spend time setting things up and fixing issues as they come up.

What makes it different

OpenClaw sits in a very specific space among AI agents.

Many direct competitors focus on outcomes without ownership. Tools like Manus AI, Lindy, O-Mega.ai, HyperWrite AI Agent, and AgentGPT are designed to be plug-and-play. You give them a task, they plan and execute it end to end. That makes them easy to adopt, but also means most of the control, logic, and data live in someone else’s system.

OpenClaw takes a different approach. Instead of offering a ready-made assistant, it gives you the infrastructure to run one yourself. You’re not hiring an agent. You’re operating it.

Compared to workflow automation tools like Zapier, Make, or Activepieces, OpenClaw is more agent-driven. Automation tools are great at connecting apps through predefined triggers and actions. OpenClaw is built around an assistant that can hold context, make decisions, and act over time, not just move data from one service to another.

There are also enterprise and niche tools like Knolli, Assembled, Atera IT Autopilot, and Jace. These are highly focused. They solve specific problems like support workflows, IT operations, or safe enterprise copilots. OpenClaw is broader and more flexible, but also less opinionated. It gives you the building blocks rather than a finished solution.

For me, what truly differentiates OpenClaw is this choice. Most agent tools optimize for convenience. OpenClaw optimises for control. You trade ease of use for ownership, flexibility, and the ability to shape how the assistant behaves over time.

That makes it a very different product from most AI agents on the market today.

My take

OpenClaw made me rethink what an AI assistant is actually supposed to be.

Most assistants today optimize for ease. You open them, ask a question, get an answer, and move on. OpenClaw doesn’t play that game. It assumes the assistant should live with you, run continuously, and take responsibility for work over time.

That choice comes with cost. Setup is heavy. You need to understand what you’re configuring and why. Things break, and you’re expected to fix them. This is not an assistant you casually adopt.

But there’s also clarity in that trade-off. OpenClaw is for people who want ownership. Who want to know where their assistant runs, how it behaves, and what it’s doing when they’re not watching. It treats AI less like a product and more like a system you operate.

For me, that’s the real takeaway. OpenClaw isn’t trying to make AI easier for everyone. It’s trying to make AI dependable for a few. And if you’re in that group, the friction starts to feel intentional rather than accidental.

It’s not the future of AI assistants for most people. But it might be the future for teams that care more about control than convenience.

AI in Design

Artlist: AI-Powered Creative Production Platform

TL;DR

Artlist combines AI image, video, and voice generation with a licensed stock asset library. It’s designed to help teams move quickly from visual direction to production-ready content.

Basic details

• Pricing: Credit based. Paid plans starting from $14/month.

• Best for: Content creators, social teams, small marketing teams

• Use cases: Marketing visuals, short videos, social content, voiceovers, quick creative iterations, production-ready assets with licensing included

Artlist is a design tool that combines licensed creative assets with AI image and video generation.

Instead of treating AI as a separate creative playground, it sits directly inside the workflow. You generate an image, tweak composition or positioning through prompts, and iterate without leaving the platform. The same environment also handles licensing and export, which keeps the focus on shipping rather than experimentation.

This makes Artlist less about expressive creation and more about practical design work. AI here functions as a fast iteration layer, helping designers move from rough visual direction to usable outputs with fewer steps.

What’s interesting

Artlist reorganizes creation around a single toolkit.

The AI Toolkit covers image generation, video generation, and voiceover, all accessible from one place. Instead of committing you to a single model, Artlist lets you choose based on the task. Models like Artlist Original 1.0 focus on cinema-grade visuals, Grok Imagine prioritizes fast, expressive images, Sora 2 Pro handles high-quality video generation, Eleven v3 (Alpha) powers voiceovers, Nano Banana Pro is geared toward detailed, consistent image outputs, and there are many more models.

These tools sit alongside Artlist’s broader catalog of music, footage, templates, and LUTs. Generated assets can be combined with licensed content and exported for commercial use without switching platforms.

And you know what’s more interesting? Artlist Studio is positioned as an upcoming AI production space. The current toolkit feels like the foundation for a more end-to-end workflow, where ideation, generation, and finishing come together in one system.

Where it works well

Artlist works well for producing marketing and content assets, not for doing detailed design work.

From my perspective, it’s most useful for content creators, growth teams, and small in-house marketing teams who need visuals, short videos, and voiceovers quickly. The value here isn’t creative depth. It’s speed and coverage across formats.

The AI Toolkit makes it easy to generate usable images and short videos in more than one way. You can start from a prompt, but you can also recreate or build on existing images and videos from Artlist’s template and asset library. That removes the blank-canvas problem and gives you a practical starting point.

I tried creating both a static image and a video. For the image, I started by generating a base using Artlist Original 1.0and then edited it using Nano Banana Pro. I was able to make a very specific change, rotating the woman so she faced the painting with her back to the viewer, without regenerating the scene or reworking the prompt. A simple instruction was enough, and building on an existing image made the process feel controlled.

I also had the option to use that image as a starting frame to generate a video. The tool clearly surfaced options to create variations and improve quality without leaving the platform.

I then tried generating a separate short video by prompting a five-second clip of a ball of wool yarn transforming into a dinosaur. The output wasn’t exactly what I was looking for, but the overall video generation flow, combined with the ability to choose between multiple models, made the process workable. Even at this stage, the result felt close to something usable for content or marketing contexts.

I find Artlist particularly effective for early visual direction. It allows me to test styles, compositions, and moods quickly, and then decide whether something is ready to ship or needs to move into a more specialised tool.

Where Artlist really delivers is in making outputs production-ready. Licensing, export, and the surrounding asset ecosystem reduce the risk and overhead that usually come with using generative AI in real projects.

Where it falls short

Artlist starts to fall short when you need more control or certainty.

I saw this while creating a short clip of a ball of wool yarn transforming into a dinosaur. The result wasn’t what I had in mind. Changing prompts or switching models meant starting again, not refining what was already there.

Some of these limits show up across the product. The templates are useful, but customization is limited. Simple changes like animation speed or text behaviour aren’t always possible. Once you move past the default look, it becomes hard to adapt.

There is also friction around usage and ownership. Assets downloaded during an active subscription can’t be used for new projects after cancellation, despite the “use forever” message. If you publish on platforms like YouTube, copyright claims can still come up. Clearing them means extra manual work.

Cost is another factor. To access music, footage, templates, and AI tools together, you need the higher-tier plan. For independent creators, that can feel expensive.

What makes it different

Artlist is different because it combines AI generation with licensed stock assets in one place.

Most all-in-one AI tools focus on one main thing. Runway is video-first. HeyGen and Synthesia focus on avatars and voice-led videos. Higgsfield AI is built for quick, mobile-first creation. These tools go deep in one area, but they’re usually designed around a single format.

Artlist takes a broader approach. It brings together AI image, video, and voice generation, alongside music, footage, templates, and export. You’re not just generating content. You’re finishing it in the same system.

Compared to platforms like Freepik AI or Envato Elements, Artlist feels more production-oriented. Freepik leans heavily into image generation. Envato offers a wide library, but its AI tools feel more like add-ons. Artlist’s AI tools are more tightly integrated into the workflow.

Against stock-first platforms like Epidemic Sound, Storyblocks, or Motion Array, Artlist pushes further into creation. Those platforms still centre discovery and templates. Artlist treats AI generation as a core part of the product.

Specialised tools like ElevenLabs, Pictory, or DomoAI go deeper in specific areas. ElevenLabs is stronger for voice. Pictory is better for turning text into videos. DomoAI focuses on stylised animation. Artlist doesn’t replace these tools, but it covers more ground in one place.

For me, what makes Artlist different is its practicality. If you’re creating across formats and care about speed, licensing, and getting something ready to ship, Artlist brings those pieces together better than most alternatives.

My take

Artlist doesn’t feel like a breakthrough AI tool. It feels like a practical production tool for the kind of content most teams ship every day.

It is not built for deep design work or precise creative control, and that becomes clear quickly, especially with video. What it does well is reduce friction. I can generate assets, make small edits, and pair them with licensed music, footage, or templates without moving between tools.

For content creators, social teams, and small in-house marketing teams, that matters. The tool helps move from an idea to something usable without requiring a lot of setup or expertise. I don’t need to overthink prompts or workflows to get to a result I can publish.

If speed and convenience are the priority, Artlist works well. If the work demands precision or originality, its limits show up fast.

In the Spotlight

Recommended watch: ClawdBot Became Moltbot...And it’s INSANE!

Julian Goldie walks through MoltBot with a grounded counterpoint to the hype around AI agents. The core idea is that most assistants still stop at conversation, while MoltBot is designed around execution. It runs locally, stays open source, and connects directly to real tools like Telegram, WordPress, browsers, Netlify, Notion, and email. Instead of building complex workflows, tasks are triggered through simple chat commands, with context and skills saved for reuse. The practical takeaway is to think of agents less as replacements for no-code tools and more as a lightweight control layer, where chat becomes the interface for automation rather than the product itself.

It’s basically what Siri should have been and it’s the closest thing I think we have to Jarvis.
– Julian Goldie • ~26:23

This Week in AI

A quick roundup of stories shaping how AI and AI agents are evolving across industries

An AI-generated Super Bowl ad for Svedka signals how generative AI is moving into high-stakes, mainstream creative work, not just experiments.

Elon Musk’s merger of xAI with X points to a push toward vertically integrated AI ecosystems that combine models, data, and distribution under one roof.

AI2’s release of Sera shows a shift toward practical, reliable coding agents built for real repository-level work, not just flashy demos.

AI Out of Office

AI Fynds

A curated mix of AI tools that make work more efficient and creativity more accessible.

Mowgli AI → An AI design copilot that turns product ideas into spec-driven, high-fidelity UI designs you can iterate on quickly.

JXP → An all-in-one AI video and image creation suite that helps you generate videos from prompts using multiple models, along with stock assets and creation tools in one place.

AI image generator → A text-to-image generation tool built for production workflows, allowing teams to generate, manage, and optimize images at scale within a broader asset pipeline.

Closing Notes

That’s it for this edition of AI Fyndings.

With Jasper helping teams turn direction into content, Moltbot exploring what it means to run an AI assistant continuously, and Artlist focusing on getting creative work out the door faster, this week highlighted a clear shift. AI tools are settling into specific roles. Less about experimentation. More about execution, control, and speed.

Thanks for reading. See you next week with more tools and patterns that show how AI is quietly reshaping work across business, product, and design.

With love,
Elena Gracia
AI Marketer, Fynd

Some shifts in AI arrive with major launches and obvious behaviour changes. Others slip in quietly, reshaping habits we already take for granted.

This edition is a little different. Once a month, we pause the weekly roundups and focus on one defining shift in AI, unpacking it with more depth and clarity.

In Perspective: This week, we’re looking at AI search visibility and what happens when search stops being a gateway and starts becoming the place where answers are resolved. As links give way to summaries and chat-based responses shape what we see first, visibility becomes less about ranking and more about how systems understand and represent information.

It’s a huge change and it’s already altering how discovery works for people, platforms, and brands alike.

Over the last year, search has started to change in a way that is easy to miss if you are only looking at traditional SEO metrics.

People are still searching. Google still handles the majority of global query volume. Rankings still update. On the surface, the system looks normal.

What has changed is where the answer now forms.

Search is no longer limited to a list of links that sends users elsewhere to think, compare, and decide. Increasingly, that work is happening inside first-party AI systems. Google’s AI Overviews and AI Mode now generate explanations directly inside the search interface. At the same time, a growing set of users start certain information and comparison queries inside tools like ChatGPT and Perplexity, where the default output is an answer, not a destination.

This does not replace traditional search behaviour. It layers on top of it.

But that layer changes how decisions are made.

When an AI system summarises options, compares tools, or explains a concept before a click happens, the role of rankings shifts. Pages can still rank. Keywords can still look healthy. And yet, fewer users click through, because the system has already done part of the cognitive work that used to happen on the website.

Recent research shows that when Google’s AI summaries appear in search results, users are far less likely to click through to websites than traditional search. In a Pew Research Center analysis of user behaviour, only 8% of visits resulted in a click to a traditional search result, compared with 15% when no AI summary appeared. And just 1% clicked links within the summaries themselves.

This is why many teams are seeing a strange pattern emerge. Rankings remain stable. Impressions stay flat or grow. Traffic quietly reduces.

Nothing is broken. The system is simply behaving differently.

Search is no longer only a gateway. It is increasingly the place where answers are formed, options are narrowed, and conclusions are reached. And once that happens, the signals teams have relied on for years begin to behave in ways that feel unfamiliar.

The shift is subtle. But its impact is not.

What AI search visibility actually is

AI search visibility is not about ranking higher on a page.

It’s about whether an AI system understands your brand, product, or content well enough to include it when it generates an answer.

Traditional search engines were designed to retrieve documents and rank them. The work of interpretation was left to the user. AI search systems are designed to produce responses. They summarize, compare, and synthesize information into something that feels complete.

That difference matters because it changes the unit of competition.

In classic SEO, you competed for placement in a list. In AI-mediated search, you compete for inclusion inside an answer. If a brand or product is not something the system can confidently interpret, it does not matter how well the underlying page ranks. It simply does not appear.

Visibility is now shifting from position to interpretation.

This is often described as a move beyond SEO, but that framing is misleading. What is emerging is a stack.

SEO still governs whether content is discoverable and trusted as a source. Systems still need to crawl, index, and understand what a page is about.

AEO, or Answer Engine Optimization, determines whether that content can be used as an answer. Can a system extract a clear explanation from the page without guessing, stitching, or misrepresenting what is being said?

GEO, or Generative Engine Optimization, sits one level above that. It shapes whether a brand appears at all in generated responses, how often it is mentioned or cited, and how it is framed across different systems.

These are not competing disciplines. You cannot do GEO without SEO, and you cannot do AEO by chasing keywords. The brands that show up consistently are not the ones producing the most content, but the ones producing the clearest sources of truth.

Where it stands now

Most companies still evaluate search performance through familiar proxies. For them, rankings indicate position, impressions signal demand, click-through rate shows whether a result acquired attention, and traffic translates to volume.

Those metrics worked because search behaviour itself was predictable. Users searched, scanned a list, clicked, and explored.

AI answers interrupt that flow.

Users still search, often in the same places, but clicking is no longer the default next step. When an answer feels sufficient, the journey ends inside the interface.

This explains why the change feels confusing from the inside. A page can rank well and still lose traffic, not because it failed, but because it was never meant to be the destination anymore. It was a source.

There is a useful parallel here with social platforms. On Instagram or LinkedIn, visibility stopped being primarily about who followed you and became about whether the platform chose to distribute you. Search is entering a similar phase. The platform is no longer only indexing information and routing attention. It is deciding when a question is considered resolved.

What gets better for people

From a user perspective, the appeal of AI-driven search is real and measurable, even if it has not fully replaced traditional search yet.

Surveys conducted with people who regularly use AI search tools find that a large majority prefer the AI experience for certain kinds of queries. In one such survey, more than 80 % of respondents said they found AI-powered search tools more efficient than traditional search engines. Not just faster, but more useful for synthesising answers and guiding decisions.

At the same time, broader population studies show that users are adopting AI search behaviour in ways that coexist with classic search engines. In a recent poll of U.S. adults, about 60% of people overall reported using AI to search for information “at least some of the time,” and that figure rose to nearly three-quarters among those under 30.

Other surveys paint a more nuanced picture: roughly a quarter to a third of users in major markets now use AI tools instead of traditional search engines for at least some queries, with younger demographics driving much of that shift.

This tells us two things that are both true right now:

1. Many users like the experience of AI search because it reduces effort and feels sufficiently complete for common decision-shaping questions.

2. A large portion of people still rely on traditional search for other query types, especially when they want to browse multiple sources, confirm accuracy, or access the latest news, and this varies substantially by age and context.

In other words, the behaviour we see is additive, not replacement at scale: people use both depending on their goals, query type, and personal preference.

AI search reduces the work people used to do manually. It summarizes long explanations, compares options without opening multiple tabs, and answers follow-up questions in the same place.

For everyday queries like:

• “best time to book flights”

• “which phone under a certain budget is better”

• “how to choose between two tools”

most people are not looking to research deeply. They want a reasonable answer and a way to move on. AI search delivers exactly that.

This is why adoption is happening quietly. People are not dramatically changing behaviour. They are simply doing less work to reach the same outcome.

It also explains why teams often feel tension in metrics. People may like the AI answer they receive because it feels efficient, but if it never results in a click to a brand’s website, traditional SEO metrics will not capture that satisfaction.

What gets worse for people

The cost of this shift shows up slowly.

When answers are generated instead of explored, the system narrows the field before users ever see alternatives. Ask an AI system for the best project management tool for a small team, and the same few products tend to appear. The framing is familiar. Strengths are summarized. Trade-offs are acknowledged just enough to feel balanced. The answer feels complete.

What stays invisible are the many specialized tools that might actually be a better fit. Not because they perform poorly, but because they are harder to summarize, discussed less often in widely trusted sources, or absent from the places the system draws confidence from.

Over time, this repetition compounds. Certain brands become defaults. Others struggle to enter the conversation at all. Discovery narrows, not through intent, but through reinforcement.

There is a second, subtler effect.

AI answers are diplomatic by design. When users ask about controversial or value-laden topics, responses tend to converge toward balance and caution. Strong critiques are softened. Adversarial perspectives lose force. The result is not necessarily incorrect information, but a narrowing of what is expressed by default.

This shapes how opinions form. Not by telling people what to think, but by quietly constraining the range of perspectives they encounter unless they deliberately push beyond the first answer.

One reason this happens lies in where AI systems learn trust.

Increasingly, that layer sits outside brand-owned websites. Community discussions, Reddit threads, YouTube explainers, Wikipedia pages, and practitioner blogs play an outsized role because they contain firsthand experience, disagreement, and concrete comparisons.

A polished brand page that avoids specifics can be difficult for a system to interpret. A messy Reddit thread full of people arguing about real trade-offs is often easier to synthesise.

As a result, brands no longer control their narrative primarily through what they publish themselves. Their visibility is increasingly shaped by what is said about them in public.

Research from AirOps adds another layer to this shift. Their analysis suggests that around half of AI search citations now come from community and user-generated sources, rather than brand-owned websites.

That means:

• brands lose control over how they are described

• the “truth layer” shifts to Reddit threads, YouTube explainers, LinkedIn posts, and Wikipedia summaries

• a strong community narrative can outweigh a polished brand website

The risk is not that AI search consistently gives wrong answers.

It’s that it gives answers that are good enough so often that users stop noticing what they are not seeing.

What it means for platforms

When we talk about platforms here, we’re referring to search and answer platforms like Google Search, ChatGPT, Perplexity, and Gemini. These systems are now converging on a similar outcome, even though they arrive there in different ways.

The moment we are talking about is this: search no longer needs to send users elsewhere to be considered complete.

For traditional search platforms like Google, this represents a clear shift in role.

For most of its history, Google’s influence came from routing attention. It decided which links appeared first, how they were presented, and where users went next. Even when it introduced featured snippets or knowledge panels, those were usually limited to simple facts. For complex or judgment-based queries, users still had to click through to other sites to finish the task.

That assumption has now changed.

With AI Overviews and AI Mode, Google increasingly resolves multi-step questions directly inside the search interface. Take a query like “how to file income tax returns.” Instead of sending users to government pages or explainer blogs, the platform now often explains the steps itself. Links still exist, but they are optional. For many users, the search ends without a click.

At this point, Google is no longer primarily deciding where you should go. It’s deciding what information is enough.

This shift also changes how platform power works.

At first, adding AI looks like deeper monopolization. Large platforms now sit closer to the moment of decision. They generate the answer, shape the framing, and often resolve the search without sending users elsewhere.

At the same time, AI weakens the old form of control.

In a link-based system, platforms monopolized attention by deciding which sites received traffic. In an answer-based system, platforms depend on a broader and more unstable source layer. AI systems pull from documentation, community discussions, videos, forums, and third-party analysis that they do not own and cannot fully control.

The result is a split.

Platforms gain power over when a search ends, but lose exclusive control over where truth comes from. Power concentrates at the level of answer formation, but fragments at the level of source creation. This is why competing systems often surface similar answers while relying on overlapping sources.

Monopolization does not disappear. It changes form.

AI-native platforms like ChatGPT, Perplexity, and Gemini arrive at this same moment differently.

They were never designed to route traffic. From the start, their role was to generate answers directly. What has changed is not their function, but their scale. As more users start their searches inside these tools, their answer-first model begins to shape discovery at a level previously reserved for search engines.

Despite different histories, both types of platforms now share the same responsibility.

They do not just organise information.
They decide when the search is over.

What this means for websites and content sources

When search was link-based, websites were destinations. Even if a page ranked lower, a compelling explanation could still win once a user arrived.

In an AI-mediated search, many websites are no longer places users visit. They are sources the system reads from.

Instead of asking, “Is this page good enough for a user to read?” AI search systems ask, “Can I confidently extract a clear explanation from this page?”

If the answer is yes, the content may be reused, summarised, or cited. If not, the page may never surface at all, even if it ranks well.

For example, imagine a user asks, “How does capital gains tax work?”

One website answers this with a long, narrative blog post. The definition appears midway through the page. The necessary information is scattered across paragraphs. Important caveats are buried near the end. The page reads well for humans, but it requires patience.

Another site answers the same question with a clear opening definition, followed by a short section explaining when capital gains apply, then a bullet-point breakdown of tax rates and exceptions. The language is plain. The structure is obvious.

In a traditional search, both pages might rank and receive traffic.
In an AI search, the second page is far more likely to be reused.

Not because it is better written, but because it is easier to interpret without error.

AI search rewards explainability over engagement.

This also elevates a specific layer of the web. Explainers, comparison pages, documentation, and community discussions increasingly shape what AI systems say, even when the brand itself is not the primary voice.

For content creators, success is no longer only about attracting clicks but about being clear enough to be used.

What it means for brands

For brands, AI search visibility introduces a new and uncomfortable reality.

Visibility is no longer just about where you rank. It is about whether you are spoken about at all, how you are described, and what role you play in the answer that forms before a click.

Rankings measure visibility in a list. Mentions and citations measure visibility in an answer.

This is why traditional SEO metrics now feel incomplete, and why observing how brands appear across AI systems has become its own layer of insight.

When people search inside ChatGPT, Perplexity, or Google’s AI experiences, the system decides which brands to mention, how to describe them, and what role they play in the answer. Often, users never see alternatives.

This is also where tools like Radix and Searchable, which I have covered in earlier editions, become relevant. They do not optimize content. They observe a different layer:

• whether a brand appears in AI-generated answers

• how often it is cited versus loosely mentioned

• how it is framed across different systems

This is not incremental SEO. It’s a new visibility surface.

What brands will do next

Once this reality sinks in, the response becomes clearer.

Brands will invest less in producing more content and more in producing fewer, clearer sources of truth. Instead of overlapping articles optimised around slight keyword variations, they will focus on canonical explanations that can travel across systems without losing meaning. Strong structure, consistent language, and explicit positioning will matter more than volume.

But clarity alone is no longer enough.

Those explanations now need to exist across multiple surfaces and formats. Text, video, documentation, comparisons, community discussions. Not because brands need to “be everywhere,” but because AI systems learn by triangulating across channels. A brand that is clearly described on its website but absent or inconsistent elsewhere is harder to include with confidence.

This shifts distribution from chasing reach to reinforcing meaning.

Brands will pay closer attention to where AI systems learn trust. Community platforms, expert commentary, and credible third-party explainers increasingly shape how brands are described, often more than brand-owned content. Being cited consistently across these surfaces becomes as important as ranking on a page.

Optimization moves away from chasing keywords and toward earning inclusion across formats, channels, and systems.

When does this become “normal”

We are already past experimentation.

AI answers are live at scale. Conversational search is being integrated into core workflows. The next phase is not adoption. It’s adjustment.

The limiting factors will not be technology. Rather, they’ll be trust, regulation, and how much decision-making users are willing to delegate without explanation.

My take

I don’t see this as the end of SEO. I see it as a correction.

Ranking still matters, but it is no longer the outcome. Being clearly understood is. Once answers are generated instead of discovered, brands compete on clarity, not volume.

What changes next is distribution.

It is no longer enough for a brand to explain itself well on its own website. AI systems learn by comparing signals across formats and platforms. If your product is clearly described in one place and vague or inconsistent everywhere else, the system will default to whatever narrative is easiest to assemble.

That means brands need to focus on being consistently citable. Across text, video, documentation, comparisons, and community discussion. Not by being everywhere, but by saying the same thing wherever they appear.

The work ahead is not chasing new optimisation tricks. It is simplifying what you publish, being explicit about trade-offs, and making your story easy to repeat without distortion.

Once search starts speaking on your behalf, the question is no longer how well you rank.

It’s whether the system can explain you correctly when you’re not there.

In the Spotlight

Recommended watch: How to Win in AI Search (Real Data, No Hype) | Ryan Law (Ahrefs)

I liked this episode because it’s one of the few AI search conversations that feels grounded in reality, not hype. Ryan makes a simple point: there are only a handful of levers you can actually influence, and most “rank in ChatGPT” advice is noise. The biggest takeaway is that mentions across the web matter more than ever. If LLMs learn what your brand is from what others say about you, then citations and co-mentions start behaving like the new backlinks. It’s also refreshingly honest about the messy part: the rules are still changing, so anyone selling certainty should make you cautious.

ChatGPT alone gets over 2.5 billion prompts every single day. And allegedly, visitors from ChatGPT are converting much better than those coming from Google
– Tim Soulo • (Intro, ~0:07)

Closing Notes

That’s it for this edition of AI Fyndings. From search shifting from exploration to resolution, to answers being shaped before we ever click, to platforms and brands adapting to a world where systems increasingly decide what information is sufficient, this week was about visibility quietly moving out of dashboards and into the logic of AI.

Thank you for reading. I hope this deep dive helped put language and structure around a shift many of us are already feeling. See you next week with more stories, tools, and ideas shaping how AI continues to change how we discover, decide, and understand the world around us.

With love,
Elena Gracia
AI Marketer, Fynd

Welcome to AI Fyndings!

With AI, every decision is a trade-off: speed or quality, scale or control, creativity or consistency. AI Fyndings discusses what those choices mean for business, product, and design.

In Business, Chronicle treats structure as the hard part of presentations, helping turn messy ideas into coherent decks before worrying about slides.

In Product, Searchable looks beyond rankings and traffic to show how brands appear inside AI-generated answers, connecting SEO work with AI-driven discovery.

In Design, Riverside focuses on workflow continuity, bringing recording, editing, and repurposing together so content doesn’t fall apart after capture.

AI in Business

Chronicle: Presentation Tool Built Around Structure First

TL;DR

Chronicle is an AI presentation tool that helps turn existing content into structured, designed decks. It handles layouts, themes, and formatting automatically, and uses AI to help shape and refine content, but still requires active involvement and review.

Basic details

• Pricing: Free plan available; paid plans start at $15 per user per month

• Best for: Tbh, anyone

• Use cases: Turning documents or prompts into structured decks, building on-brand reports, strategy decks, updates, and sales pitches

Most of us default to PowerPoint or Slides not because they help us think, but because they’re familiar.

They are the tools we’ve always used. You open a template, start adding slides, and figure out the placement as you go. Even when the goal is clear, the workflow pulls you into mechanics too early. You are adjusting layouts, rewriting headings, and moving boxes around while trying to express ideas you already understand. Slides get added to support the message, but the structure can start to obscure it.

Chronicle offers multiple ways to start, but the intent remains consistent. You bring in whatever you already have, and the tool helps turn it into a structured presentation. Layouts, themes, and formatting are handled in the background, which shifts more of your attention to shaping the content instead of arranging slides.

What’s interesting

There are four ways you can start working with Chronicle: paste text, start with a prompt, import content from a website, PDF, or PPT, or begin with a template.

I tried Chronicle in two ways. I uploaded a PDF, and I used the prompt-based flow.

I used a mock strategy document with dummy information when I uploaded the PDF. I expected Chronicle to convert it into slides and largely preserve the content as-is. That didn’t happen. Instead, it treated the PDF as raw input. It broke the content into sections, reorganised it, and rewrote parts of it to fit a presentation format. Some sections were condensed, others were reframed.

The prompt-based flow was interesting in a different way. I asked Chronicle to help create a deck on “Digital Marketing Trends Report: 2026 Outlook for AI and SaaS Leaders.” Instead of generating slides immediately, it asked questions first. Who is this for? What’s the angle? What should someone take away from it?

To test it further, I deliberately pushed it in multiple directions. In response, it pushed back. It suggested narrowing the scope instead of creating a long, unfocused deck.

This is not something I usually see with AI tools like Gamma and Beautiful.ai. Most of them either agree or expand whatever you give them. What I liked about Chronicle was how it reshaped messy input, whether it was a rough document or a broad prompt, into something more coherent before moving into the design stage.

Where it works well

Chronicle works well when you are building or refining a presentation.

The editor gives you flexibility without forcing you to design everything manually. You can add different types of content directly into a slide or chapter, including text, media, cards, embeds, diagrams, sticky notes, and connectors. This makes it easier to structure information visually without designing every slide from scratch.

It also provides ready-made deck templates across use cases like strategy, planning, reports, pitches, and reviews. You can start with one to get structure quickly, or build from scratch and apply structure later. You can use preset themes or create your own brand theme by defining colours, fonts, and styling. Once a theme is created, it can be saved and reused across presentations, which keeps decks consistent without extra effort.

You can also use the AI throughout the process. You can ask it to generate slides, expand a section, or rework content at any point, instead of being limited to the starting step.

Where it falls short

Chronicle is useful, but it’s not something I’d trust on autopilot.

The AI helps, but it isn’t consistent enough to rely on without checking. Some sections come out fine, others feel generic or miss context, especially when the input is complex. I ended up reviewing and editing most of what the AI generated.

Imports are also uneven. If you’re pasting content with tables, dense text, or uneven structure, the layout doesn’t always translate cleanly. You often have to tidy things up manually.

There are gaps in the workflow as well. If you refresh the page or start a new session, you can’t return to the same AI conversation. There’s no chat history, which makes it harder to iterate over time.

Chronicle is also not very practical for editing on smaller screens or making quick changes on the go.

And if you’re working on larger decks, you can run out of tokens quickly unless you’re on a paid plan.

What makes it different

Chronicle treats structure as the core problem. That’s also where its trade-offs show up.

If you already know what you want to say and just want to turn it into slides quickly, other tools can feel faster. Tools like Gamma and Tome optimize for speed and flexibility. You give them a prompt and they generate something immediately, even if the output needs reshaping later.

Chronicle is more restrictive, but that restriction prevents things from sprawling too easily.

Compared to Beautiful.ai, Chronicle offers more control over structure but less visual polish by default. Beautiful.ai makes slides look good with minimal effort. Chronicle’s output is clean, but it relies more on the quality of the underlying content to land well.

Against tools like Pitch or traditional slide editors, Chronicle feels less mature. Pitch has stronger collaboration and integrations and fits more easily into existing workflows. Chronicle’s approach is more opinionated, which can be useful, but also makes it harder to slot into every team setup.

My take

I find Chronicle useful when I’m still figuring things out.

If the content is messy and the structure isn’t clear, it helps slow things down and organise thinking before turning it into slides. That’s where it adds value.

But if I already know exactly what I want to say and just need slides quickly, I’d probably reach for something like Gamma or Beautiful.ai.

AI in Product

Searchable: SEO Tool Built for an AI-Driven Search World

TL;DR

Searchable is an SEO and AI visibility platform that helps teams understand how their content performs both on traditional search engines and inside AI-generated answers.

Basic details

• Pricing: No free plan. Paid plans start at $50 per month

• Best for: SEO, product marketing, growth, and content teams adapting their search strategy to include AI-driven discovery

• Use cases: Auditing site and content quality, identifying keyword and content gaps, tracking brand visibility in AI answers, understanding competitor presence across SEO and AEO

Search has started to change, but most product and marketing teams are still measuring it the old way.

Traffic, rankings, and keywords made sense when people searched through links. Now, a growing share of discovery happens inside AI tools, where users get answers directly and brands either show up or don’t, without a click ever happening.

Searchable helps teams understand how their brand appears inside AI-generated answers across tools like ChatGPT, Perplexity, and Google’s AI results. Instead of focusing only on rankings and traffic, it tracks prompts, citations, and visibility inside AI responses, and then helps teams act on that information.

What’s interesting

I’ve looked at tools in this space before, including Radix, which focuses specifically on tracking how brands appear inside AI-generated answers. Radix is useful for answering a narrow question: are we being mentioned, and where?

Searchable approaches this from a wider angle.

I tried Searchable by running a homepage audit for Kaily.ai, an AI agent that goes beyond the chatbox.

Instead of starting with AI mentions, the product starts with the basics. It breaks the page down into technical health, content quality, and AI visibility, and then ranks issues by severity. That structure made it easier to see what actually mattered, rather than treating every issue as equally urgent.

The feedback was also more concrete than I expected. When Searchable flagged the value proposition as unclear, it didn’t just label it as a problem. It explained what the page currently communicates, why that message doesn’t land clearly, and how it could be rewritten to be more direct.

For keyword analysis, instead of listing keywords in isolation, it grouped them by intent and use case. Each group came with context around why it mattered and what kind of content would realistically support it. That made the output feel closer to a content plan than a keyword report.

Where Searchable really differs from Radix is in how it treats AI visibility. Instead of showing mentions in isolation, it connects prompts, pages, competitors, and sources. You can see which queries surface your brand in AI-generated answers across OpenAI, Perplexity, and Google’s AI results, and which competitors show up instead, with clear links back to why that might be happening.

What stood out to me is that Searchable doesn’t frame AI discovery as a separate, future-facing problem. It treats it as a downstream outcome of content quality, structure, and distribution, building on familiar SEO workflows rather than replacing them.

Where it works well

Searchable is strong at bringing different signals together in one place. Prompts, visibility trends, mentions, competitors, sources, issues, and recommendations all live inside the same workflow. You don’t have to jump between tools to understand what’s happening and what to do next.

It’s also useful for ongoing monitoring. Once prompts, competitors, and domains are set up, you can track changes over time without rerunning everything manually. Visibility trends, share of voice shifts, and new mentions become easier to spot. This makes it helpful not just for one-time audits, but for keeping an eye on how things evolve.

Searchable works well for product marketing, growth, SEO, and content teams who are already doing search or planning content and want to understand how that effort translates into AI-driven discovery. It’s less about replacing existing workflows and more about adding a layer of clarity on top of them.

Where it falls short

Searchable brings a lot of information into one place, but it takes time to understand how everything connects.

Compared to Radix, Searchable is more complex and slower to fully understand. Radix is easier to adopt if your primary goal is monitoring AI mentions and citations. You can get value from it quickly without needing to interpret multiple layers of data.

Searchable, on the other hand, asks you to do more synthesis. It brings audits, keywords, prompts, competitors, sources, and recommendations into one place, but it doesn’t always tell you exactly what to do next. You still need experience and judgment to decide which issues are worth acting on.

Some of the gaps Searchable highlights are also not fully in your control. AI systems rely heavily on third-party sites, forums, and comparison pages. While Searchable makes these dependencies visible and explains their impact, improving your position often requires work outside the tool and over a longer time horizon.

Where I land is this: Radix is clearer and faster if you want a focused view of AI mentions. Searchable is more useful if you want to understand how SEO, content quality, and AI visibility connect. The trade-off is simplicity versus context.

What makes it different

Searchable is built around the idea that SEO and AI visibility are no longer separate problems.

Most SEO tools still focus on rankings, traffic, and clicks. Newer AEO tools, like Radix, Otterly, and Promptwatch, focus on whether your brand appears in AI answers. Searchable tries to connect these two worlds instead of picking one.

Instead of stopping at rankings or keywords, it shows how that SEO foundation translates into visibility inside AI-generated answers. You can see which prompts surface your brand, which competitors appear instead, and which sources AI systems rely on.

Compared to pure AEO tools like Otterly, Promptwatch, or Peec AI, Searchable goes further upstream. It doesn’t just monitor mentions. It tries to explain why those mentions exist and what you could change on your site or in your content to influence them.

My take

Searchable is useful when you’re trying to understand how your SEO work shows up in an AI-driven world.

It helps connect familiar things like content quality, keywords, and technical health with a newer question: whether your product is actually part of AI-generated answers. That makes it more practical than tools that only track mentions, and more forward-looking than tools that only track rankings.

For me, Searchable is the stronger choice over Radix. It’s more comprehensive and gives me enough context to move from insight to action. I’d use it to decide what to fix, what to create, and where to focus next, especially as AI-driven discovery becomes increasingly hard to ignore.

AI in Design

Riverside: Recording and Production Tool for Podcasts and Video Content

TL;DR

Riverside is a recording and production platform that helps creators and teams record high-quality audio and video, then turn those recordings into publishable content in the same workspace.

Basic details

• Pricing: Free plan available; paid plans start with a 14 day trial and then $29 per month

• Best for: Podcasters, video creators, and marketing teams producing content regularly

• Use cases: Recording remote podcasts and interviews, editing audio and video, creating clips and captions, repurposing long recordings into short-form content

If you’ve worked on a podcast or a marketing video, you know recording isn’t the hardest part. The real work begins after.

Editing the episode. Pulling clips. Adding captions. Reformatting for different platforms. What starts as one recording quickly turns into a long production checklist.

Riverside is built for that entire workflow.

It lets you record high-quality audio and video with remote guests, and then edit and prepare that recording for publishing in the same place. You can cut and arrange the recording, create clips, add layouts and branding, generate captions, and export different versions without moving files between tools.

Instead of treating recording, editing, and design as separate steps, Riverside brings them together in one product. You record once and use that recording to create multiple pieces of content.

What’s interesting

What stood out to me about Riverside is that once a session ended, I didn’t have to move to another tool or figure out what to do next. The recording stayed in the same workspace, already laid out and ready to edit. That continuity made the whole process feel lighter than I expected.

I found the text-based editing especially useful. Instead of scrubbing through a timeline, I could edit by cutting words directly from the transcript. For longer conversations, this made cleaning things up much faster and less tiring.

I also liked the Magic Clips feature. I didn’t need to export sections and rework them elsewhere. I could highlight moments, turn them into short clips, change layouts, add captions, and adjust framing, all within the same flow.

Where it works well

Riverside works well when you’re creating content regularly without a dedicated production or editing team.

It’s especially useful for teams that need to move quickly after recording. Being able to clean up audio, edit through text, create clips, add captions, and apply basic branding in the same place saves time. I didn’t have to pass files between tools or rethink the workflow for each output.

Riverside is also a good fit for marketing teams that need one recording to travel across multiple formats. Long-form episodes, short clips, social videos, and branded snippets can all come from the same session without starting from scratch each time.

Where I found it particularly helpful was in reducing handoffs. Producers, hosts, and editors can work in the same environment, which keeps things simpler. You don’t need a full production setup to get something out the door.

Where it falls short

Riverside works well for many common workflows, but it has clear limits.

The editor is useful, but it doesn’t replace a full editing tool. I could do basic trims, cleanups, captions, and clips comfortably, but for more detailed audio or video work, it starts to feel restrictive. If you’re used to deeper control over edits, you’ll eventually want to move parts of the work elsewhere.

The AI features help speed things up, but they still need review. Captions, summaries, and clips are a good starting point, not a finished output. I had to check and adjust most of what the AI generated before using it.

Recording itself is generally reliable, especially with local recording, but the experience isn’t completely seamless. Bringing in guests, managing connections, or handling edge cases can still introduce small interruptions during sessions, even if the final files are preserved.

What makes it different

Riverside feels different when you compare it to the tools people actually reach for when they need remote recording and production, not just one part of the workflow.

Compared to SquadCast or Zencastr, which focus primarily on capturing remote audio and video, Riverside goes further. With those tools, you record and then export files to work on them elsewhere. With Riverside, the recording stays in the same workspace where you edit and prepare content. You record once, and when the session ends, you’re already in a production environment rather than facing a separate handoff.

Compared to Descript, the difference shows up in how the product is oriented. Descript is an editor-first tool, with recording as one part of the workflow. Riverside is capture-first. Everything else builds out from the recording session, and the AI tools are woven into that flow instead of feeling like separate modes.

For me, what makes Riverside different is this sense of workflow continuity. Capture, edit, and first-pass production all happen in one place. You don’t record in one tool, edit in another, generate clips in a third, and caption somewhere else. It’s all connected.

My take

Riverside is useful when you’re creating content regularly and don’t want the process to fall apart after recording.

It helps you move from a finished recording to usable content without jumping between tools. Editing, clips, captions, and basic design all happen in one place, which makes the workflow easier to manage.

It’s not meant for heavy editing or creative experimentation, and that’s fine. What it does well is help teams turn recordings into publishable content consistently.

I’d rely on Riverside regularly and not something I’d look at occasionally.

In the Spotlight

Recommended watch: SEO vs AEO in 2026: The New Rules for Winning on Google and AI Models

Ethan Smith breaks down a calmer view of the “SEO is dead” narrative. His core point is that traditional search traffic has stayed roughly flat, while LLM-driven discovery is growing fast but remains much smaller than Google. The practical takeaway is to treat SEO and answer engine optimization as one blended strategy, with a few AEO-specific moves like targeting long, specific questions and earning off-site mentions in the exact URLs that show up in citations.

The pie is increasing in size. It’s not that it’s shifting from search to LLMs. It’s that there’s an increased usage of both.
– Ethan Smith • ~5:08

This Week in AI

A quick roundup of stories shaping how AI and AI agents are evolving across industries

Meta launches AI Glasses Impact Grants, a program funding research and projects that use AI-powered glasses for social and assistive applications.

Spotify rolls out AI-prompted playlists for Premium users, enabling playlist creation through natural-language prompts in the US and Canada.

OpenAI publishes its approach to advertising, explaining how ads could support broader access to its AI products while prioritising user trust and safety.

AI Out of Office

AI Fynds

A curated mix of AI tools that make work more efficient and creativity more accessible.

Rewarx → An AI product photography and virtual studio that turns simple snapshots into high-quality 4K images and commercial video assets for ecommerce and marketing.

PDF Beauty → An AI-powered PDF to editable PPTX converter that rebuilds static slides into fully editable PowerPoint files while preserving layout and design.

AI Video Generator → An AI video generation tool for automating video creation and adapting assets across workflows.

Closing Notes

That’s it for this edition of AI Fyndings.

With Chronicle slowing teams down to clarify structure before slides, Searchable helping teams understand how their work shows up inside AI-driven discovery, and Riverside keeping recording and production in one continuous flow, this week pointed to a common theme. AI is increasingly shaping the process around the work, not just the output, guiding how ideas are formed, refined, and carried through.

Thanks for reading. See you next week with more tools, patterns, and ideas that show how AI is steadily changing how we work across business, product, and design.

With love,
Elena Gracia
AI Marketer, Fynd

Welcome to AI Fyndings!

With AI, every decision is a trade-off: speed or quality, scale or control, creativity or consistency. AI Fyndings discusses what those choices mean for business, product, and design.

In Business, Octolane tries to reduce CRM work by following sales activity where it already happens, in email and calendar.

In Product, Firecrawl uses AI to sit between intent and code, helping teams explore web data before committing to full scraping pipelines.

In Design, Spacely focuses on iteration over output, letting designers work through a space instead of generating a final image upfront.

AI in Business

Octolane: CRM That Tries to Keep Up With Sales

TL;DR

Octolane is an AI-first CRM that reads activity from Gmail and Calendar to update deals and draft follow-ups automatically. Reduces manual CRM work, but still needs close review and doesn’t fully replace traditional systems.

Basic details

• Pricing: Free plan with paid plans starts from $49/per seat/month

• Best for: Sales teams that operate primarily through email and calendar

• Use cases: Creating automatic deals, drafting follow-up emails, keeping CRM data updated with minimal manual input

When was the last time your CRM actually reflected what was happening in your inbox?

That doesn’t usually happen. Deals move forward in emails and meetings, but the CRM lags behind unless someone goes back and updates it manually. That gap isn’t dramatic, but it compounds. Follow-ups get delayed. Pipelines lose accuracy.

Octolane helps deal with that lag by connecting to Gmail and Calendar, tracking your activity, updating records in the background, and drafting follow-ups for review. Instead of treating the CRM as a place you visit, it treats it as something that stays in sync with how you already work.

What’s interesting

What I found interesting about Octolane is that instead of opening on a dashboard, it used my login to analyze my emails and calendar and understand the deals I was already working on.

Once that analysis was done, a lot of information was already in place. Accounts and contacts were laid out clearly, and deals came with details like ownership, close date, probability, recent interactions, and next steps already filled in. Not all of it was accurate, but it reduced the amount of manual work I would otherwise have had to do.

The email drafts also stood out. Octolane picked up context from past conversations and wrote follow-up emails based on what had already been discussed. I only needed to review the draft once before sending it directly from the tool.

It also shows deal predictions. I wouldn’t rely on them, but they give a sense of where the product is headed and how it’s trying to stay closer to day-to-day sales work.

Where it works well

Octolane works best for small sales teams where Gmail and Google Calendar are the primary systems of work, including teams that actively close and lock in deals through ongoing email and meeting-based conversations.

It’s not fully accurate, and some details still need to be checked and corrected, but the overall effort is much lower than maintaining everything by hand.

For now, it works better for simpler sales setups. If your deal stages and fields are straightforward, the tool is easier to use. Teams with complex workflows or heavily customized CRMs may find it limiting.

It works best when used as a tool that prepares information, not something you rely on blindly.

Where it falls short

The biggest issue for me was accuracy. It doesn’t get the basics right. In my case, it created a deal with the wrong contact name, which immediately made me lose interest.

It also didn’t pull in all deals. After the analysis, many conversations and opportunities were missing, which meant the view it created wasn’t complete or accurate. That made it hard to rely on the tool to reflect everything I was working on.

I constantly felt the need to monitor it. I couldn’t depend on it blindly. I had to review the data, fix details, and sometimes add information myself, which takes away from the idea of a fully self-running CRM.

There’s also no clear review or approval flow. While drafts and data are generated quickly, there isn’t a structured way to manage reviews or validate what’s been created, especially if more people were to use it.

A sales manager I spoke to, Ritesh, rated it 2 out of 5 and described it as a basic tool. That feedback matched my experience. The idea is strong, but the execution still feels early.

What makes it different

Octolane stands out for how it updates a CRM. Most CRMs, including tools like Attio, focus on structured fields such as deal stage, value, and next steps, and depend on users to update those fields manually. Lightfield focuses on reading conversations to build context. Octolane takes a different approach. It looks at email threads and calendar meetings and tries to create or update deals, contacts, and follow-ups based on that activity.

It is built around automatic updates, not suggestions. This matters because the AI is not just recommending what to do next. It is actively creating records, moving deals forward, and drafting follow-ups using what it sees in emails and meetings.

This design makes Octolane feel less like a tool you operate and more like a system that runs in the background. Instead of asking you to log activity after the fact, it tries to capture sales activity as it happens and reflect it directly in the CRM.

My take

I like the direction Octolane is taking. Letting the CRM follow email and calendar instead of forcing people to work inside it makes sense, and when it works, it genuinely reduces some of the busywork around sales.

But right now, I don’t trust it enough to run on its own. The accuracy issues, missing deals, and lack of a proper review flow mean I have to stay involved. That limits how much time it actually saves.

I’d use Octolane as a supporting tool, not as my main CRM. It’s helpful for keeping track of conversations and reducing manual updates, but it still feels early. The idea is solid. The product needs more time to catch up.

AI in Product

Firecrawl Agent: An AI Shortcut Between Intent and Scraping Code

TL;DR

Firecrawl Agent is an AI-powered tool for extracting structured data from public websites using prompts and schemas instead of custom scraping logic. It works well for exploration and prototyping, but struggles with scale, limits, and reliability.

Basic details

• Pricing: Free plan available with paid plans starting from $16/month

• Best for: Developers, data teams, and product teams working with public web data

• Use cases: Web scraping and crawling, structured data extraction in JSON or text, early experimentation and data validation

A lot of modern software work depends on data that lives on the open web. Product catalogs, pricing pages, listings, documentation, and content sites are all public, but not easily usable. Getting that data usually means writing custom scrapers, handling JavaScript-heavy pages, and fixing things every time a layout changes.

Firecrawl’s Agent steps into that workflow. It’s a tool designed to extract structured data from websites without forcing you to handcraft the entire scraping logic yourself. You describe what you want, point it to a URL, and choose the output format, while the system handles crawling, rendering, and extraction in the background.

What’s interesting

What’s interesting about Firecrawl’s Agent is how it turns web extraction into a guided process instead of a one-shot scrape.

When I ran it on the Reliance Digital website, it didn’t assume what I wanted. It asked clarifying questions, let me define the structure, and then worked toward producing data in that shape. The interaction felt more like configuring a task than writing scraping logic.

I also liked the schema-based output. Being able to define fields upfront and receive the results directly in JSON made the output immediately usable. There was no extra step of cleaning or restructuring the data after extraction, which usually takes up a lot of extra time.

Where it works well

Firecrawl works well for tasks where the goal is to extract data from public websites without building and maintaining custom scraping logic.

In my experience, it was most effective for early exploration and one-off extraction tasks, where I needed to quickly test whether a site could yield usable data. I didn’t have to invest upfront time in writing or debugging code just to validate the idea.

Based on this, Firecrawl feels well suited for developers, data teams, and product teams who regularly work with public web data and want a faster way to experiment, prototype, or support internal tools. It’s especially useful when speed of setup and visibility into the process matter more than accuracy or scale.

Where it falls short

The biggest issue I faced was reliability during the first few runs. On the Reliance Digital website, the extraction did not work well initially. I had to try multiple times, change the prompt, and adjust the schema before I got usable results. This kind of trial and error is manageable for testing, but it makes the tool harder to trust for consistent use.

The token limit is another major constraint. I hit the 200k token limit very quickly, even while testing. When I tried to extract more products or run a broader crawl, the job failed. This makes it difficult to test or work with larger websites without breaking tasks into smaller parts.

I also noticed that you need to be very specific when working with bigger websites. Vague prompts don’t work well. The more complex the site, the more detailed the instructions need to be. On smaller or simpler websites, the tool works more smoothly and requires less effort.

The tool also only works with public websites. Anything behind login screens, paywalls, or restricted access is out of scope, which limits where it can be used.

What makes it different

What makes Firecrawl different isn’t that it scrapes the web. Plenty of tools already do that, and many of them do it better.

The difference is how Firecrawl uses AI.

If I compare it to tools like Selenium, Beautiful Soup, or Scrapy, those are still far more reliable once they’re set up. They’re deterministic. You write the logic, handle edge cases, and know exactly what will happen. The downside is that everything is manual. Any change on the website means more code and more maintenance.

On the other end, platforms like Apify or Diffbot handle scale and stability much better. They’re built to run large jobs repeatedly. But they also hide a lot of the process. When something breaks, you often just see a failed job, not why it failed.

Firecrawl takes a different path. It uses AI to replace hard-coded scraping logic with intent. You describe what you want, define the structure, and the agent tries to figure out how to get there. It makes decisions along the way, retries when things fail, and adapts its approach. Crucially, you can see all of this happening.

That visibility is what stands out. Firecrawl doesn’t pretend the web is clean or predictable. It shows you the mess. But that also means you’re relying on AI judgment instead of fixed rules, which introduces uncertainty.

So the trade-off is clear. Firecrawl is faster to start with and easier to experiment on, but it’s less stable and less predictable than mature tools. It’s not replacing traditional scraping or managed platforms. It’s changing how much effort you put in before you get something usable.

Whether that’s a good thing depends on how comfortable you are trading control for flexibility.

My take

Using Firecrawl changed how I think about web extraction more than how I’d actually do it today.

What it showed me is that AI can sit between intent and code. I didn’t have to decide upfront how a site should be scraped. I could start with what I wanted and see how far the tool could take me. That makes early exploration faster and less intimidating.

But it also made the limits very visible. The more complex the site, the more specific I had to be. The more data I wanted, the more fragile the process became. At that point, the AI stopped feeling like leverage and started feeling like something I had to manage.

So I don’t see Firecrawl as a replacement for scraping tools. I see it as a bridge. It helps you move from “can this work?” to “this is what would be required.” That’s a useful place for a product to exist, even if it’s not where the work ultimately ends.

For me, its value is less in the data it returns and more in the clarity it gives you early on.

AI in Design

Spacely: AI-Assisted Interior Design Iteration

TL;DR

Spacely is an AI-assisted interior design tool focused on editing and iterating on existing spaces. It supports localized changes, lighting control, and visual comparison, making it useful for early design exploration rather than final production.

Basic details

• Pricing: Free plan available with paid plans starting from $15/month

• Best for: Interior designers, architects, and visualization teams

• Use cases: Early-stage design exploration, visual iteration and comparison, client presentations and internal reviews

Most interior design and architecture work is visual, but a lot of the effort goes into producing visuals rather than exploring ideas.

Trying out a different lighting setup, adding a piece of furniture, or changing a wall detail usually means reworking a render or regenerating an image entirely. That makes small design decisions slower than they need to be.

Spacely is an AI design tool built for working on existing spaces. It allows designers to make targeted changes, compare versions, adjust lighting, and add or remove objects without starting over each time.

What’s interesting

Spacely allows you to work on a specific space without forcing you to regenerate everything each time.

I started by working on an existing interior rather than generating a new image from scratch. From there, I could make very specific changes. Using Point & Edit, I clicked on a particular area of the image and added a traditional bohemian painting to the room. The painting appeared exactly where I clicked, and the rest of the space stayed the same.

For object-level changes, I used Auto Edit and Remove & Replace. With Auto Edit, I could describe a change in simple language, like adding a folded knitted throw near the cushions, and the tool applied that change to the image. With Remove & Replace, I could remove an existing object or replace it with something else. In both cases, the changes were limited to that object and didn’t affect the rest of the design of the room.

What’s more was I could control the lighting of the room. Instead of treating lighting as a single style choice, it lets you work with it in detail. I could switch between lighting presets like daylight or evening light, and I could also adjust individual light sources. Each light could be changed by color, intensity, radius, and position. This made it possible to test different moods without changing the layout or materials.

Spacely also lets you work on architectural elements like the ceiling. I could design and modify the ceiling independently. That kind of control is usually hard to get in AI-based interior tools.

I also explored the Collage mode, which works differently from prompting. Instead of asking the AI to generate objects, I could place 2D or 3D assets directly into the space from an asset library. I could move them around, resize them, and build the scene piece by piece. This felt more manual and controlled, with AI supporting the process rather than taking over.

In addition to this, Spacely offers style and space presets, which make it easier to explore different design directions easily. You can switch styles or room types without starting from scratch, which helps when you want to test multiple options.

Across all these features, the AI is embedded into specific actions instead of acting like a chatbot. The tool guides you on how to write prompts, limits the scope of changes, and keeps edits focused. What stood out to me is that Spacely lets you work in small steps, change one thing at a time, and keep refining the same space instead of committing to a final image early.

Where it works well

Spacely feels best for exploring different variations of a space rather than producing a final, locked-in render.

I feel it’s most suited for early design exploration and iteration. Basically for when you want to test multiple designs quickly, it will help you go back-and-forth without having to redesign everything each time.

It’d fit well into internal reviews, concept discussions, or early client conversations.

Where it falls short

Spacely focuses on visual exploration more than technical designing. You are only working around with images, and the controls are based on visual adjustments instead of precise measurements.

There are multiple modes and tools and it takes time to understand how to approach them.

What makes it different

When I compare Spacely to other tools in the design and visualization space, what stands out isn’t a single “better feature,” but a different way of working.

Visualization tools like Twinmotion and Lumion help designers with high-quality, production-ready renders, but they require significant setup, scene preparation, and design expertise. They are good, but they are also technical and time-intensive. Spacely doesn’t replace these tools, and it doesn’t aim for production-grade realism or interactive environments. What it brings instead is speed of iteration without heavy technical overhead.

On the other hand, generative AI image tools like Midjourney, DALL-E, or Stable Diffusion can create inspiring visuals from prompts, but they treat output as single, final images. If you want to make a specific change, like moving a lamp, adjusting lighting, or editing only the ceiling, you often have to regenerate the whole image or rely on workarounds.

In my experience with Spacely, the ability to make targeted edits using Point & Edit, Auto Edit, or Remove & Replace without disturbing the rest of the room is something prompt-only tools don’t offer.

There are also design-centric AI platforms that lean toward full automation and try to generate a complete style solution from minimal input. Spacely doesn’t try to guess the design. Instead, it lets you stay in control and make small decisions step by step, which feels closer to how designers actually work.

So the difference isn’t that Spacely does everything better. It sits in a middle ground. It is less technical and heavy than full visualization engines, more editable and controlled than prompt-only generative tools, and less automated than end-to-end design automation platforms.

My take

What stayed with me while using Spacely was the way it let me stay inside the design process instead of jumping in and out of it.

I didn’t feel like I was “generating images.” I felt like I was working on a space. Making a small change, seeing how it looked, adjusting something else, and continuing from there. That’s not something most AI design tools get right today.

This is what I feel product managers and designers building AI tools should pay attention to. Spacely doesn’t optimize for impressive outputs. It optimizes for flow. It keeps the user oriented, lets discussions build on each other, and avoids forcing a reset every time you want to explore a new idea. This pattern should apply far beyond interior design.

At the same time, I was always aware of what it is and what it isn’t. Spacely doesn’t replace professional design tools, and it’s not meant for final or technical outputs. It’s a visual thinking tool. It helps you explore, explain, and iterate, especially when ideas are still forming.

In the Spotlight

Recommended watch: The Future of Architecture: A look beyond the AI hype

Olly Thomas (Design Tech at BIG) gives a grounded view of AI in architecture: it’s not “idea magic,” it’s a force-multiplier for iteration, visualization, and the boring admin work. The upside is simple: less time rendering and formatting, more time on judgement, research, and design thinking.

The value of what we do as architects… is still the thinking… that hasn’t changed.
– Olly Thomas • ~7:43

This Week in AI

A quick roundup of stories shaping how AI and AI agents are evolving across industries

Anthropic introduces Claude as a coworker, emphasizing persistent context, tool use, and ownership of real work, not just answers.

Slack turns Slackbot into a context-aware AI agent, embedded directly in conversations and workflows where work already happens.

Google releases the Universal Commerce Protocol (UCP), an open standard that enables AI agents to browse and transact across merchants, pushing commerce toward an agent-first model.

AI Out of Office

Source: Kaily AI

AI Fynds

A curated mix of AI tools that make work more efficient and creativity more accessible.

AnimateAI → An AI tool that turns text prompts into short animated videos. Useful for quick motion concepts, explainers, or lightweight social content without a full animation setup.

Meshy → An AI-powered 3D generation tool that creates models from text or images. Useful for rapid 3D ideation, asset exploration, and early-stage spatial thinking.

AI Video Generator → An AI video generation tool designed to automate video creation and transformations. Useful for producing and adapting video assets across product and marketing workflows.

Closing Notes

That’s it for this edition of AI Fyndings.

With Octolane trying to reduce the overhead of keeping CRMs updated, Firecrawl helping teams explore web data before committing to full pipelines, and Spacely making design iteration more fluid, this week pointed to a shared shift. AI is moving closer to the actual work, not by replacing people, but by taking on the repetitive effort that slows progress down.

Thanks for reading. See you next week with more tools, patterns, and ideas that show how AI is quietly reshaping how we work across business, product, and design.

With love,
Elena Gracia
AI Marketer, Fynd

Welcome to AI Fyndings!

With AI, every decision is a trade-off: speed or quality, scale or control, creativity or consistency. AI Fyndings discusses what those choices mean for business, product, and design.

In Business, Munch Studio shows what happens when social media shifts from a task you manage to a system that runs.

In Product, Final Round AI explores how interview tools move beyond preparation into live support.

In Design, Meta’s SAM 3D examines what changes when selection becomes the starting point for working in 3D.

AI in Business

Munch Studio: Social Media Without Weekly Decisions

TL;DR

Munch Studio is an AI tool that plans, creates, and schedules social media content end to end. Built to keep social media running without needing constant attention, even when no one owns it full-time.

Basic details

• Pricing: Paid with pricing starting at $48/month

• Availability: Web-based

• Best for: Businesses without a dedicated social media manager

• Use cases: Social media planning, content creation, scheduling, and publishing

One thing I keep seeing across teams is this: social media is not hard because it’s complex. It’s hard because it’s constant.

There’s always something to decide. What to post. Which platform to focus on. Whether a reel or a carousel makes more sense. How often is “enough.” What’s on brand. What’s trending. What’s working. What’s not.

These decisions keep coming back every week, and most teams don’t have a clear system for making them. So progress depends heavily on whoever is paying attention at that moment.

Munch Studio is built to replace that recurring decision-making with a preset system.

Munch is essentially an AI tool that takes your brand, your website, and your inputs, and then turns that into a full social media setup. It decides the platforms, creates a content strategy, plans the week, generates posts and videos, and schedules them. You can guide it but the default behaviour is that the system keeps moving without waiting for you to decide each step.

It isn’t built as just a creative assistant, but as a system that tries to keep your social media running.

What’s interesting

What stood out to me first is how much Munch does with very little input. A website link and basic brand details are enough for it to put together a full setup. Platforms, themes, posting frequency, and a weekly plan appear quickly, without needing prompts or detailed instructions.

Another interesting part is how clearly it shows its thinking. You can see the goals, the audience it’s targeting, the content pillars, the keywords, and how all of that translates into posts for the week. Nothing feels hidden. Even if you don’t agree with the decisions, you can clearly see what the system is operating on.

I even liked how the content isn’t generated randomly. It’s organized by platform and format. The posts, stories, reels, carousels, and videos are planned separately, with a defined number of posts per week. It feels less like “make content” and more like “manage output.”

While Munch directly parses your brand using just your website, you can also add your brand guidelines, create a brand kit, upload your own media, and adjust tone. Those inputs are consistently applied once the system is set up.

You can plan, create, schedule, and publish all in one place. You can also connect your Instagram, Facebook, LinkedIn, YouTube, and TikTok, publish directly, and view basic performance metrics.

It’s interesting how everything is connected into one system that keeps running once it’s set up.

Where it works well

Munch works best when social media needs to exist, but shouldn’t take up too much thinking time.

From what I saw, this setup makes sense if the problem isn’t the lack of ideas, but the effort it takes to keep deciding what to do every week. Once it’s set up, Munch removes that loop. Platforms, formats, posting frequency, and schedules are already decided, so the system keeps moving.

I think this works particularly well for small teams. If no one owns social media full-time, this kind of structure helps. Since everything sits in one place, it becomes easier to manage other priorities.

And even if your brand’s social media is inconsistent or inactive, it helps lower the effort needed to start again. You don’t have to make every decision upfront.

Where it falls short

The strategy Munch creates is fine as a base, but it doesn’t go very far. It follows common patterns. If I were looking for a strong point of view or a clear story for a brand, this wouldn’t be enough on its own.

Secondly, the creative quality is inconsistent. Some posts make sense. Some don’t. Captions are okay, but they lack human voice. I wouldn’t be comfortable publishing everything without reviewing it first.

Even with brand inputs in place, the nuance can get lost. A distinct voice doesn’t always come through, especially if the brand relies on tone or personality more than visuals.

The performance data is also limited. I can see basic numbers, but there’s no real explanation of what’s working and why. It tells me what happened, not what to do next.

For me, this means Munch reduces effort, but it doesn’t remove responsibility. I’d still want someone involved to review, adjust, and course-correct.

The bigger risk isn’t bad content, but content that’s just good enough to go unnoticed.

What makes it different

Most AI social media tools fall into three buckets.

Some tools focus on scheduling. Tools like Buffer or Hootsuite help you publish content once you already know what to post. They don’t help with planning or strategy.

Some tools focus on content creation. Tools like Ocoya help generate captions, creatives, and videos. They speed up creation, but you still decide what goes out and when.

Some tools combine creation and scheduling in one place. They reduce tool switching, but the planning still sits with you.

Munch takes a different approach.

Instead of helping with parts of the workflow, it tries to run the whole thing. It creates a strategy, chooses platforms, plans posting frequency, generates content, and fills the calendar. Planning and execution happen together.

That’s the key difference. Munch is not just helping you post or create faster. It’s taking over the weekly decisions most teams struggle to keep up with.

The tradeoff is control. You give up some flexibility in exchange for consistency.

My Take

After using Munch, my takeaway is simple.

This is not a tool that makes social media better. It makes social media easier to keep running.

If someone expects strong ideas, a clear voice, or thoughtful storytelling, Munch won’t deliver that on its own. The output needs checking and direction.

But if the real problem is that social media keeps slipping because no one has the time or patience to manage it every week, then this kind of system makes sense.

I don’t see Munch as a replacement for a social media manager or a creative team. I see it as a safety net. Something that keeps the channel running.

Whether it’s useful or not depends on what role social media plays for you.

If it’s core to the brand, this isn’t enough.
If it’s important but not central, it probably is.

AI in Product

Final Round AI: Interview Support Across Preparation and Performance

TL;DR

Final Round AI is an interview-focused platform that combines resume building, interview practice, live interview support, and job applications.

Basic details

• Pricing: Free plan available; Pro plan required for live interview copilot with prices starting at $90/month

• Availability: Web-based with Chrome extension for live interviews

• Best for: Job seekers, early- to mid-career professionals, candidates interviewing remotely

• Use cases: Resume building, mock interviews, live interview assistance, job applications

You know that moment in an interview when your mind goes blank, not because you don’t know the answer, but because everything hits at once.

You’ve prepared. You know the example. But the question is phrased slightly differently, your attention splits between listening and thinking, and suddenly you’re trying to respond in real time. By the time you find your footing, the conversation has already moved on.

Final Round AI is built for that exact situation. Instead of focusing only on preparation, it supports candidates during the interview itself. Using your resume, role context, and interview setup, it helps structure responses as the conversation unfolds.

From a product perspective, this is the real shift. The interview is treated as an active workspace, not a moment where tools disappear. That decision defines how the product is built and where it places its value.

What’s interesting

I tried Final Round AI by setting up a mock interview using my resume and role, and then exploring how the interview copilot works once an interview begins.

What stood out is how focused the product is on the interview itself. Most of the experience is built around helping you answer questions as they come up, not just preparing you beforehand or summarising things later.

You can control how the AI responds while the interview is happening. You can change how detailed the answers are, which language they’re in, and how structured they feel. You can also choose formats like Default, STAR, or SOAR. These controls are easy to access and actually change how the answers sound.

Your resume stays active throughout the product. It’s used during mock interviews, live interview answers, and even when applying for jobs. You don’t have to keep re-explaining your background. The system carries that context forward.

Even the mock interviews are simulated very close to the real interview experience. The settings are the same for both.

There’s also the AI Job Hunter feature, which lets you set preferences, see matched roles, and apply quickly or in bulk. For now, it’s a little limited in terms of locations and roles but I hope to see it expand soon.

Where it works well

From what I saw, Final Round AI works best when interviews feel harder to handle than to prepare for.

Suppose you’ve been asked to walk through your work experience. You know the answer but might struggle to explain it clearly on the spot. In such a case, Final Round AI goes through your experience and gives your answer a simple structure. It helps shape responses while you’re speaking, instead of leaving you to figure that out mid-answer.

This is especially useful in behavioural interviews. These interviews care more about clarity than speed. Final Round AI helps keep answers focused, so they don’t trail off or miss key points.

It also works well if you’re interviewing with multiple companies at the same time. When interviews pile up, it’s easy to repeat the same examples badly or mix things up. Because the tool always pulls from the same resume and role context, answers stay consistent across interviews.

I also think this is helpful for early- and mid-career candidates. These are people who may have the experience but aren’t yet confident talking about it. The tool supports how answers are framed without taking over the conversation.

Finally, it works for people who want everything in one place. Resume building or uploading, practice interviews, live interview support, and job applications all sit inside the same system, which makes the process easier to manage.

Overall, I see Final Round AI working best when interviews feel overwhelming not because of lack of skill, but because there’s too much happening at once.

Where it falls short

Final Round AI is helpful when interviews follow a clear structure. It is less effective when conversations become highly technical, fast-moving, or unstructured. In those situations, the suggestions can lag behind the discussion or feel too generic to be useful.

The live interview copilot also depends on specific setup conditions. You need a Chrome extension and access to the Pro plan.

The job search feature of the product is still limited. It currently supports only a narrow set of job roles and locations, and global job listings are not yet available.

Finally, the tool does not remove the need for human judgement. Responses often need review and adjustment, especially in interviews that require precision, nuance, or strong personal perspective.

What makes it different

If you look at interview tools today, most of them do one thing.

Tools like Google Interview Warmup or Interview Sidekick help you practise answers. They’re useful before the interview, but they stop there.

Tools like Parakeet AI help you improve responses quickly, but they’re still limited to preparation. They don’t stay connected once the interview actually starts.

There are also tools like Cluely that sit slightly differently. It focuses on what happens during meetings and interviews by helping you keep track of conversations, notes, and context in real time. It supports recall and clarity, but it doesn’t shape or structure what you say.

Whereas, Final Round AI tries to handle more of the interview journey in one place. You can work on your resume, practise interviews, get support during the interview, and apply to jobs from the same product. Most tools don’t attempt that range.

Compared to simpler tools, Final Round AI feels heavier. You need more setup, and key features sit behind a paid plan. Compared to tools like LockedIn AI, which focus almost entirely on live interview assistance, Final Round AI feels less specialised.

So Final Round AI isn’t the best at one specific thing. What makes it different is that it tries to connect everything together. Whether that’s a good thing or not depends on what you’re looking for.

If you want a focused tool for one job, there are simpler options.
If you want one system to cover most of the interview process, Final Round AI is trying to do that.

My take

Final Round AI brings several parts of the interview process into one place. Resume building or uploading, interview practice, live interview support, and job applications all sit inside the same product.

From what I’ve seen, the live interview copilot is the most developed part of the experience. Other areas, like job discovery and role coverage, still feel limited in comparison.

I think this works best for people who want a single system to manage interviews end to end, rather than switching between multiple tools. It’s less compelling if you’re looking for depth in one specific area.

Whether it’s useful depends on how much you value having everything connected versus using specialised tools.

AI in Design

SAM 3D: Exploring 3D Through Simple Interaction

TL;DR

SAM 3D is a research demo from Meta that turns selected objects in images into rough 3D models. It lets designers move from a flat image to basic spatial understanding without heavy 3D setup.

Basic details

• Pricing: Free research demo

• Availability: Web-based demo via Meta’s Segment Anything gallery

• Best for: Designers, researchers, and creators exploring 3D and spatial ideas

• Use cases: Object segmentation, early 3D exploration, spatial prototyping, research experimentation

Working with visual assets breaks down the moment you move beyond flat images.

Designers can isolate objects in 2D fairly well today, but the moment something needs to exist in 3D, the workflow becomes slow and fragmented. You either need multiple images, specialised tools, or manual modelling. Even simple objects take time to recreate, and the process rarely fits into everyday design work.

SAM 3D extends Meta’s Segment Anything work from 2D selection into 3D reconstruction. Instead of starting with complex 3D pipelines, it begins with a simple interaction designers already understand: selecting an object in an image. From a single image, SAM 3D can generate a rough 3D representation of objects or human bodies that can be inspected and manipulated in space.

The tool is part of Meta’s broader effort to make visual understanding more flexible across images, video, audio, and now 3D. From a design perspective, SAM 3D is not positioned as a replacement for traditional 3D tools, but as a way to make spatial assets easier to create and experiment with earlier in the workflow.

What’s interesting

I tried SAM 3D through Meta’s Segment Anything demos, starting with images and then exploring the other tools in the gallery.

The first thing that stood out was how easy it is to get started. You upload an image and click on the object you care about. SAM highlights it immediately. If the selection isn’t right, you fix it using simple Add and Remove clicks. There’s no setup or tuning.

Once the object is selected, you can generate a 3D version of it from that same image. The result isn’t perfect, but you can rotate it, inspect it, and understand its shape in space. It feels more like a quick way to explore an idea than a final output.

What I found interesting is that SAM 3D sits alongside several other tools that use the same basic idea. In the gallery, you can:

• search for objects in video and see them tracked across frames

• turn a person in an image into a simple 3D body with pose information

• isolate or remove specific sounds from audio by describing what you want

Even though these are very different tasks, the interaction stays almost the same. You either point at something or describe it, and the system tries to separate it for you. The interface doesn’t change much across image, video, 3D, or audio.

Taken together, these tools feel less like finished products and more like experiments. Once you can reliably pick out what you want to work with, everything else becomes easier to build on top of it.

Where it works well

SAM 3D works well as an exploration and prototyping tool, not a production one.

It’s useful for designers who want to quickly pull objects out of images and think about them in 3D without setting up complex tools. If you’re sketching ideas, testing layouts, or trying to understand how something might exist in space, this makes that step much faster.

This is especially useful for:

• Product and interaction designers experimenting with spatial or 3D ideas

• 3D designers and artists looking for quick reference shapes before rebuilding assets properly

• Researchers and technologists exploring how selection and perception can work across media

• Creators working with mixed formats like images, video, and audio

It also works well for early-stage workflows. Instead of spending time cutting, masking, and rebuilding assets, you can get something usable in minutes and decide whether it’s worth taking further.

Where it falls short

The 3D output is rough. You can look at it and rotate it, but it’s not something you can use as-is. Shapes feel incomplete, and details are often missing. Any serious use would still need proper 3D work afterward.

It also struggles with messy and detailed images. When objects overlap or the background is noisy, the segmentation isn’t always clean. Even with Add and Remove clicks, it can take effort to get a usable selection.

There’s very little control beyond the selection itself. You don’t get options to improve mesh quality, fix textures, or guide how the 3D model is built. You mostly need to accept whatever the system produces.

Some of the other demos feel early. Video tracking can break when objects move fast. Audio separation works only in simple cases and with precise prompts. The body-to-3D demo gives you a generic result that’s far from realistic.

What makes it different

Tools that turn images into 3D are mostly based out of a few specific categories.

Some tools use photogrammetry. Software like 3DF Zephyr needs many photos taken from different angles to create a 3D model. The results can be accurate, but the process is slow and requires a lot of setup.

Other tools rely on phone-based scanning. Apps like Qlone let you scan objects using your camera, but you still need the right lighting, positioning, and physical space to get usable results.

There are also newer AI-based approaches that generate depth or multiple views from a single image. These are useful for visualisation, but they often do not produce a clean 3D model you can actually work with.

SAM 3D attempts to go straight from a single image to a 3D object. You select what you want in the image, and the system predicts its shape and texture in 3D. There is no need for multiple photos or special hardware.

This makes SAM 3D much faster and easier to try than traditional 3D tools. At the same time, the results are rougher and less precise. You gain speed and simplicity, but you give up detail and control.

Another difference is how SAM 3D fits into a larger system. It builds on Meta’s Segment Anything work, so selecting an object works the same way across images, video, and even audio. Most 3D tools separate selection, capture, and modelling into different steps and tools.

My take

SAM 3D is not a tool for creating finished 3D assets.

What it does well is help you understand shape and space quickly. You select something in an image and get a rough 3D version almost immediately. That makes it easier to think in 3D without setting up complex tools.

The output is limited and rough, so this isn’t something you’d use for production work. But it’s useful for early exploration, references, or testing ideas.

In the Spotlight

Recommended watch: AI, Algorithms & A Social Media Marketing Playbook

This is worth watching because it explains why marketing feels harder and faster at the same time. We’ve moved from social media to interest media, where algorithms decide what people see based on behaviour, not who they follow. In that world, distribution is fluid and attention shifts quickly. AI becomes essential not for automation, but for keeping up. It helps marketers sense what’s changing, test ideas faster, and stay relevant in feeds that reset every day.

I think we’re fully in interest media now. When I open my app, I get stuff I’m into of the moment. And the second I start to deviate, that starts to show up.”

– Gary Vaynerchuk, ~4:14-4:20

This Week in AI

A quick roundup of stories shaping how AI and AI agents are evolving across industries:

Elon Musk’s AI chatbot Grok on X has been widely misused to generate non-consensual, sexualised images of women and children, drawing global backlash and regulatory scrutiny over AI safety and content moderation.

Gmail enters the Gemini era, embedding AI directly into the inbox to help users write, summarise, search, and manage email as an ambient layer of everyday work rather than a separate assistant.

OpenAI launches ChatGPT Health, positioning ChatGPT as a clearer, more trustworthy interface for understanding medical information while signalling a careful push into high-stakes, regulated domains.

AI Out of Office

AI Fynds

A curated mix of AI tools that make work more efficient and creativity more accessible.

Music Video Generator → Turn a music track into an AI-generated video by choosing a visual style and letting the model handle the visuals.

Patterned.ai → Create seamless, repeatable patterns with AI for backgrounds, packaging, and brand assets.

AI Background Remover → Instantly remove image backgrounds in the browser for clean product and marketing visuals.

Closing Notes

That’s it for this edition of AI Fyndings. With Munch keeping social media running without constant attention, Final Round AI supporting interviews as they happen, and Meta’s SAM 3D making early 3D exploration easier, this week was about AI stepping into parts of work that usually slow us down.

Thanks for reading! See you next week with more tools, shifts, and ideas that show how AI is quietly changing the way we work.

With love,
Elena Gracia
AI Marketer, Fynd

Welcome to AI Fyndings!

With AI, every decision is a trade-off: speed or quality, scale or control, creativity or consistency. AI Fyndings discusses what those choices mean for business, product, and design.

In Business, Motion turns execution into a scheduling problem. It records meetings, captures action items, and continuously replans tasks against real calendars so commitments stop living in limbo.

In Product, Trickle makes tool building feel lighter. It turns prompts into working interfaces, and gives you a canvas where pages, data, assets, and rules stay connected as you iterate.

In Design, Topview treats short-form ads as a repeatable format. Instead of designing every video from scratch, it uses proven ad structures and platform patterns to generate usable creatives fast.

AI in Business

Motion: Turning Plans Into Scheduled Work

TL;DR

Motion is an AI-powered work planning tool that turns tasks, meetings, and deadlines into a continuously updated schedule. It helps individuals and teams follow through on commitments by planning work against real calendar availability.

Basic details

• Pricing: 7-day free trial. Paid plans starting at $49/mo for individuals and at $29/per seat/mo for teams.

• Best for: Can be used by individuals and teams alike

• Use cases: Task scheduling, deadline management, meeting follow-ups, execution planning

Most teams don’t struggle with knowing what needs to be done. They struggle with fitting that work into the time they actually have.

Tasks live in one place. Meetings fill up calendars elsewhere. Deadlines sit somewhere in between. When priorities change (which they always do), planning becomes manual and reactive. Work still gets tracked, but there’s no reliable system translating it into actual time. Important tasks don’t get dropped because they’re forgotten. They get dropped because they’re not actively re-planned.

That gap is what Motion is trying to solve.

It brings tasks, projects, meetings, and availability into one system and uses AI to schedule work directly onto your calendar. Instead of asking users to decide what to work on next, Motion makes that decision based on deadlines, priorities, and real capacity.

And when your plans change, for example a meeting gets rescheduled or a new urgent task is assigned, Motion adjusts the schedule rather than leaving it on you to reshuffle everything yourself. Tasks are treated as time-bound commitments, not static checklist items.

When I first set it up, I connected my calendar right away and started using it for personal planning. It immediately picked up my existing schedule and gave me an option to add tasks. There was very little friction in getting started. It didn’t feel like I was configuring a complex workflow or teaching the tool how I work. It was adapting to the constraints that already existed.

Within the first day, it felt like this approach could actually hold up in practice. The interface is simple to navigate. Your work doesn’t just get organized. It gets realistically planned and scheduled.

From a business perspective, Motion uses AI for execution discipline. It doesn’t generate ideas or content. Its value lies in continuously translating work into time and making trade-offs visible before deadlines are missed.

What’s interesting

What stood out to me about Motion was how much ownership it took after a meeting ended.

I ran a mock meeting (which quickly escalated into a real one) where I asked my manager to assign me tasks. After the meeting ended, Motion sent a detailed MOM email to all attendees. Inside the tool, the full meeting recording was available, along with structured notes.

More importantly, the tool had already picked up on what was assigned to whom. The tasks discussed in the meeting were converted into action items, assigned to the right people, and scheduled directly into calendars based on deadlines and availability.

I didn’t have to write notes, extract tasks, or decide when the work should happen. There was a brief moment of hesitation when I saw tasks placed at times I wouldn’t have chosen instinctively. But the logic was clear. The schedule was grounded in real calendar constraints, not arbitrary guesses.

What made this compelling wasn’t the meeting recording itself. It was how accurately Motion understood the conversation. It picked up proper nouns, responsibilities, and timelines without prompting, and turned them into planned commitments.

Being able to ask the system what I should work on next and get a concrete, time-aware answer reinforced what Motion is optimised for. It’s not about capturing meetings. It’s about closing the gap between discussion and execution.

Where it works well

From my experience, Motion works best as a personal execution system for project-driven work.

I found it most natural for managing project tasks where deadlines exist and work needs to be planned across a busy calendar. When tasks are clear and time-bound, Motion does a good job of turning them into a realistic plan without requiring constant manual reprioritisation.

Even in personal use, the structure makes it clear how Motion would scale to team contexts. It plans tasks against real calendars and visible availability, which is especially useful when tasks have dependencies or when multiple people are working toward the same outcome. In setups like shared projects or team-managed deliverables, this kind of time-aware planning can reduce coordination overhead.

Another area where Motion works well is consolidation. Tasks, meetings, notes, and docs all live in one place. This doesn’t dramatically change how you think about work, but it reduces friction. There’s less context-switching and less reliance on memory to connect planning, documentation, and execution.

Motion also suits roles that sit somewhere between ideation and execution. For people who spend time thinking, planning, and then handing work off or following through on it, the tool helps bridge that gap by making execution explicit. It’s equally relevant for individuals managing their own projects and for those responsible for keeping teams aligned on shared work.

Overall, Motion works well for people and teams who already know what needs to be done and want a system that helps them deliver it consistently.

Where it falls short

Motion’s main limitation shows up when its assumptions don’t line up with reality.

There were moments when the timelines Motion proposed felt unrealistic. Tasks were scheduled in ways that technically fit the calendar but didn’t always reflect how work actually unfolds. In those cases, I had to step in and make adjustments manually.

This isn’t a major flaw, but it highlights an important boundary. Motion relies on the inputs it’s given. When task durations or constraints are imperfect, the resulting plan can feel overly optimistic. It is good at working within time, but it can’t fully account for nuances like energy levels or task complexity. In those cases, you still need to step in and adjust things manually.

The mobile experience is more limited than the desktop version. It works well for checking your schedule or reviewing tasks, but it is not ideal for making detailed changes on the go.

Beyond that, I didn’t run into significant friction. Even simple, everyday tasks fit comfortably into the system, and it handled a mix of work without feeling restrictive or heavy.

Overall, Motion’s limitations are less about missing features and more about calibration. It works best when timelines are realistic and expectations are clear. When those inputs drift, the schedule needs human correction.

What makes it different

Motion sits in a crowded ecosystem of work tools, and its value only became clear to me once I stopped thinking about what it competes with and started looking at what it actually takes ownership of.

Before using it, I instinctively compared Motion to tools like Jira or meeting note-takers like Granola. After spending time with it, that comparison stopped making sense.

Jira is designed to track work across teams. It’s good at workflows, visibility, and progress, especially for delivery-focused teams. But tasks largely live in an abstract state. They move through statuses, not through actual time on a calendar. From my experience, Jira tells you what exists, not when it will realistically happen.

Granola operates much earlier in the flow. It’s built around meetings. It helps capture discussions, decisions, and context so nothing gets lost. That’s useful, but it stops at understanding. It doesn’t deal with execution.

Motion, in contrast, assumes decisions are already made and focuses almost entirely on what happens next.

What felt different in practice was how opinionated Motion is about execution. Tasks aren’t just listed. They are placed into time alongside meetings, based on deadlines and availability. When there isn’t enough time, that constraint becomes visible immediately instead of being hidden in a backlog.

Unlike meeting tools, Motion doesn’t help you think through ideas or make sense of conversations. It helps you follow through. It takes outcomes from discussions and turns them into scheduled commitments that have to fit into real calendars.

The AI is not the headline feature here. What stood out to me is how narrowly it’s applied. It exists to continuously rebalance plans as reality changes, not to generate content or suggestions for their own sake.

Motion doesn’t replace tools like Jira or meeting note-takers. It fills the gap they leave behind. It’s the layer that turns decisions into work that actually gets done.

My take

Motion is best understood as an execution tool, not a general productivity app.

It works well when work is already defined and deadlines matter. If tasks are clear and time-bound, Motion helps by planning them against real calendar availability and keeping that plan updated as things change. This reduces the need for constant manual planning and reshuffling.

A large part of Motion’s value comes from having everything in one place. Tasks, meetings, notes, and schedules live together, which makes execution feel more concrete. The meeting notetaker supports this by ensuring decisions don’t stay abstract, but the real benefit is how consistently those decisions turn into planned work.

Motion does require a certain level of clarity to be effective. Tasks need reasonable durations and deadlines, and calendars need to be kept accurate. When that information is off, the schedule becomes less trustworthy and needs human correction.

The tool is less helpful for work that is still taking shape. It assumes decisions have already been made and struggles when tasks are open-ended or hard to scope. That isn’t a flaw so much as a design choice.

Overall, Motion works best for individuals and teams who care about predictable delivery and follow-through.

AI in Product

Trickle: Turning Prompts Into Product-Shaped Experiments

TL;DR

Trickle helps you turn product ideas into working interfaces using AI. It’s useful for early exploration, internal tools, and proofs of concept where seeing something real matters more than finishing a production build.

Basic details

• Pricing: Free plan available, paid plans starting at $20/mo

• Best for: Product managers, designers, small teams

• Use cases: Early product exploration, proofs of concept, quick experimentation

After Lovable, it’s clear this category isn’t going away. The question is which part of the problem each tool actually solves.

In one of our earlier editions, I spoke about how Bubble represents the control-heavy end of this category, offering deep flexibility at the cost of setup. Lovable sits at the other end, prioritising speed by generating working apps quickly, but treating that output as a starting point rather than a workspace you keep returning to. Both aim to reduce the cost of building products, but they optimise for very different trade-offs.

Trickle sits in that same space, but approaches the problem differently.

Instead of focusing on setup or one-off generation, Trickle is built around a persistent canvas. You describe what you want to build, and Trickle generates a working web interface backed by real code. That output stays live inside the product, where you can preview it, edit it, and keep iterating without starting over.

This changes how early product work feels. You are not wiring things together from scratch, and you are not constantly regenerating disposable versions. You are working on something that already behaves like a product, while it is still easy to adjust and reshape.

That positioning is what makes Trickle worth paying attention to in this category.

What’s interesting

The most interesting thing about Trickle is that it doesn’t treat “prompting” as the product. It treats the canvas as the product.

In Trickle, the unit of work is a Magic Canvas: a visual workspace where pages, assets, rules, data, and notes sit together, and the AI builds inside that context. It’s not just “generate me a page” and hope it remembers. You can literally attach constraints (rules/notes), drop in assets, add business knowledge, and then ask it to expand or refine from there. The canvas becomes the memory.

That structure changes how iteration feels. Instead of regenerating from scratch, you keep evolving the same artefact: add a page, tweak layout, change styling, inject docs, roll back a version if it goes sideways. Trickle even supports version history and restore, which makes experimentation less risky.

The other underrated bit: Trickle is clearly trying to be end-to-end, not “a builder + three other tools.” Inside the same workflow, it can:

• generate and manage image assets (and save them into a library for reuse)

• create a built-in database for dynamic content (AI can set it up, or you can prompt it)

• publish live directly from the canvas

And yes, this is where the “coding” angle shows up. Trickle positions the canvas as something that’s continuously translated into production-ready code, and it also lets you download/export code files (with the important caveat that exports don’t include database data and don’t carry over Trickle’s AI functionality).

So what’s interesting about Trickle isn’t that it can generate websites. A lot of tools can do that. It’s that Trickle is trying to make the workspace smart: a single place where intent, context, assets, logic, and pages live together, and the output keeps getting better because the inputs are actually organised.

Where it works well

Trickle works best when you’re trying to understand whether an idea holds up once it becomes real.

In my case, I tried two very different builds inside Trickle: a small tape recorder utility and a basic e-commerce shopping website. Both helped clarify where the product is genuinely useful.

The tape recorder app is a good example of how Trickle handles small, focused tools. Trickle generated a complete interface in one go, with clear controls and a sensible layout. It wasn’t a design mock or a loose concept. It behaved like a simple utility you could actually imagine using. For internal tools or small features, that’s often enough to evaluate whether something is worth taking further.

The e-commerce shopping website showed the same strength at a larger scale. Trickle created a recognisable storefront structure with navigation, product listings, and page hierarchy. Seeing everything together made it easier to reason about the overall flow than looking at isolated screens. This is especially useful early on, when the goal is to understand structure and direction rather than polish.

In both cases, Trickle handled iteration well. Changes could be made without breaking the underlying structure, which made it easier to explore variations without starting over.

Publishing adds another practical layer. Being able to put these outputs on a custom domain makes them easy to share for feedback or internal review. It turns experiments into something people can interact with, not just look at.

In terms of who this is for, Trickle works best for founders, product managers, and designers who want to validate ideas quickly with something tangible. It also suits small teams building internal tools or early proofs of concept. For larger organisations, it fits best for experiments, internal utilities, or side projects rather than core production systems.

Where it falls short

Trickle gets expensive fast if you’re using it heavily.

On the free plan, usage is capped by daily credits (and only up to a monthly cap). Once you’re iterating a lot, you feel the ceiling. Same with the built-in database: the free plan only supports 100 rows, so anything slightly real (like e-commerce data) hits limits quickly.

Publishing also has constraints.

You can publish, but custom domains aren’t available on free, and even paid plans have a defined number of custom domain connections. If you want the output to feel “real” outside Trickle, you end up needing Pro/Premium.

Exporting code is useful, but it’s not a clean handoff.

Trickle lets you download code files, but the export does not include database data and does not include any AI functionality tied to the project. So if you’re thinking “I’ll build here and then move it to a normal dev workflow,” there’s a gap.

Finally, it still demands precision from you.

Trickle’s own guidance is basically: the more specific your prompts are, the better the results. In practice, that means vague prompts don’t hold up. You have to write clearly, iterate patiently, and sometimes steer it like a teammate, not a tool.

What makes it different

To understand where Trickle fits, it helps to look at how similar tools approach building.

Tools like Webflow or Bubble give you a lot of control, but they expect you to build everything step by step. You design layouts, wire logic, and structure pages manually. You get flexibility, but it takes time before anything usable exists.

On the other hand, tools like Durable or Wix’s AI builder are very fast. You get a website or app almost instantly. But once it’s generated, making meaningful changes often means regenerating large parts of it. Iteration feels shallow.

Lovable sits closer to Trickle, but with a different focus. Lovable is strong at quickly generating app structure and behaviour together. It gives you a fast starting point with logic and data flows already implied. You move quickly, but the experience is still centred around generation first, then adjustment.

Trickle takes a different route. It treats the canvas itself as the core product, not just the output. You generate once, then keep building inside the same workspace. Pages, assets, data, prompts, and changes all stay connected. Instead of repeatedly generating new versions, you evolve the same one.

Compared to AI coding tools like Replit AI or Cursor, Trickle hides the code. You don’t need to write or debug anything directly. That makes it easier to use if you’re not a developer, but it also means you’re not working at the same level of control as a full coding environment.

In simple terms:

• Trickle is slower than one-click generators, but more flexible once you’re inside

• It’s faster than traditional no-code tools at the start, but not as deep in the long run

• It’s easier than AI coding tools, but less precise

What makes Trickle different is that it’s designed for building and iterating inside one workspace, rather than jumping between tools or regenerating outputs. That choice shapes both its strengths and its limits.

My take

Trickle is most useful when you want to turn an idea into something tangible without setting up a project or writing code.

The tape recorder app and the e-commerce site made this clear. In both cases, Trickle helped me move past abstract thinking and see how an idea would actually look and feel as a product. That alone changes the quality of decisions you can make early on. It’s easier to spot gaps, unnecessary features, or wrong assumptions when you’re clicking through something real.

Where Trickle fits best for me is early product work. Exploring ideas, building internal tools, testing flows, or aligning with others in direction. It shortens the distance between “we think this could work” and “this is what it would look like if it did.”

I would not use Trickle for building a production system. Once requirements become strict, logic becomes complex, or long-term ownership matters, the trade-offs become obvious.

AI in Design

Topview: Designing Short-Form Ads Without Designing Them

TL;DR

Topview helps turn product assets into short-form video ads that follow proven platform patterns. It’s built for speed, consistency, and repeatability rather than creative exploration.

Basic details

• Pricing: Free plan available, paid plans starting at $29/mo with credit-based usage

• Best for: Marketers, small teams, e-commerce brands

• Use cases: Short-form product ads, performance marketing creatives, TikTok and Instagram Reels, rapid creative testing

Short-form video ads have become the default format for product discovery, but producing them at scale is still messy.

Teams need videos that follow platform conventions, open with a strong hook, reflect brand guidelines, and still feel native enough to not get skipped. Doing this repeatedly means scripting, storyboarding, sourcing footage, editing, adding text and music, and exporting in the right formats. It’s fast work, but it’s rarely simple.

Topview is built to simplify that workflow.

Instead of treating video creation as a single prompt, Topview breaks it into steps that mirror how ads are actually made. You start with product images and a brief. The system extracts the main details, defines a structure, generates a storyboard, and then produces a finished video designed for short-form platforms like Instagram Reels and TikTok.

What makes this interesting is how deliberate the process feels. Topview behaves less like a creative tool and more like an execution layer for ad production. The AI isn’t just generating visuals. It’s making decisions about pacing, structure, and format based on common patterns in high-performing videos.

Topview is clearly optimised for speed and repeatability. It’s not trying to replace high-end video production or creative direction. It’s trying to make producing usable, platform-ready ads much easier to scale.

What’s interesting

Topview stands out for how much creative structure is embedded directly into the product.

The Viral Video Agent is not a one-time generator. It keeps context across iterations, which means you can generate variations or change direction without starting over. This makes iteration feel more controlled and reduces the need to repeat the same inputs again and again.

The agent templates add another layer of structure. Formats like viral hook, POV, shoppable skit, or ASMR are not just labels. Each template follows a different pacing and narrative pattern, and that shows up clearly in the final video. Choosing a template directly shapes how the video opens, how the product is introduced, and how the message flows.

The storyboard view supports this approach. Before the final video is generated, Topview shows a clear sequence of scenes with timing and intent. This makes it easier to review the flow and catch issues early, instead of reacting only after the video is rendered.

Brand consistency is handled through the Brand Kit. You can set logos, end cards, fonts, music, AI voices, and avatars once, and reuse them across videos. These settings carry over, which makes repeat ad creation more consistent without extra setup.

The Ad Library adds useful reference context. You can explore ads by region, industry, engagement, or estimated revenue. This brings inspiration and benchmarking into the same place where videos are created.

What’s interesting about Topview is that many creative decisions are already built into the tool. Instead of starting from a blank page each time, you’re working within a system that guides the structure of the output.

Where it works well

Topview works well when the goal is to produce short-form product videos quickly and repeatedly.

The skincare ad example shows this clearly. Starting with product images and a brief, Topview generated a complete video with a clear hook, product focus, benefit highlights, and a closing frame. The output followed common short-form ad patterns without needing manual editing or sequencing. For performance-driven content, that structure is useful.

Topview is especially effective for repeatable ad formats. Once a product is set up, generating multiple variations becomes straightforward. You can try different templates, hooks, or styles without rebuilding the video from scratch each time. This works well for testing creatives across platforms like TikTok and Instagram.

The tool also fits teams that need speed over polish. For small marketing teams or solo founders, producing videos in-house can be time-consuming and expensive. Topview reduces that overhead by handling scripting, pacing, and editing in one place. You don’t need separate tools for storyboarding, editing, and exporting.

For businesses, Topview makes the most sense in performance marketing and e-commerce contexts. It works well for brands running frequent campaigns, launching new products, or refreshing ad creatives regularly. The Brand Kit helps maintain consistency across videos, which matters when ads are produced at scale.

At an individual level, Topview is useful for creators, indie founders, and small sellers who want professional-looking product videos without a production setup. It lowers the skill and time required to produce ads that match platform norms.

Topview is less about creative exploration and more about execution at scale. It works best when the objective is to produce usable, platform-ready ads quickly, rather than crafting highly bespoke or cinematic videos.

Where it falls short

The biggest constraint is that Topview is credit-driven, so heavy iteration adds up quickly. A single Viral Video Agent export costs credits, and Avatar 4 model (where you can create lifelike characters using a photo and script) is charged per second. If you generate a few variations in one sitting, you feel the burn.

Video length is also capped by plan. The Video Agent max length is 15 seconds on Free and Pro, and 25 seconds on Business. That’s fine for many short ads, but limiting if you want slightly longer product explainers.

The free plan is genuinely usable, but it comes with clear restrictions. You get 10 one-time credits, limited access to the TikTok Ad Library, and exports include a watermark. That’s enough to test the workflow, but not enough to run it as a consistent production pipeline.

Even on paid plans, credits have rules. For example, Pro and Business credits are issued yearly and expire annually. So it’s not an “unlimited use” feeling product. You’re still budgeting usage.

Rendering speed is also part of the trade-off. Topview explicitly positions faster rendering as a paid-plan benefit, and Avatar 4 has modes (including scheduled/off-peak behaviour). In practice, this means speed is not always constant across usage choices.

Topview is strong at producing short, platform-shaped ads. The friction shows up when you want to iterate a lot, go longer than short-form, or rely on it daily without thinking about credits and plan limits.

What makes it different

Most AI video tools fall into one of two buckets.

Tools like Canva, CapCut, or InVideo are built around editing. They offer templates and timelines, but you still decide how the video is structured, how the story flows, and how scenes are paced. AI helps with assets and suggestions, but the creative decisions remain manual.

On the other end, tools like Synthesia or HeyGen focus on avatars and narration. They work well for presenter-led videos, explainers, or internal communication. However, the output is usually linear and informational, not shaped like a performance ad.

Topview takes a different approach.

It is built specifically for short-form product ads, not general video creation. The tool assumes you are trying to make something that looks and behaves like a TikTok or Instagram ad, and everything is designed around that goal.

Instead of asking you to design the flow, Topview assembles it for you. The hook, product focus, benefit sequence, and closing are inferred by the system. You are not editing a timeline. You are choosing intent and letting the tool put the video together.

Topview also differs from avatar-first tools. Avatars and voices are available, but they are not the centre of the experience. The focus stays on product visuals, motion, text overlays, and pacing that matches how ads perform on short-form platforms.

Another distinction is the built-in Ad Library. It is not only a source of inspiration. It shows real ads by region, industry, and engagement, which helps set expectations for what works in practice. This keeps outputs aligned with existing patterns rather than pushing toward experimental formats.

In short, Topview is not trying to help you edit videos or explore creative ideas. It is designed to help you produce ads that already fit platform norms, with fewer decisions along the way. That narrow focus is what makes it different and also defines where it is most useful.

My take

Topview is a tool designed for a very specific job, and it works best when used that way.

It’s effective at turning product assets into short-form ads that already follow familiar platform patterns. The tool removes much of the manual decision-making around structure, pacing, and format, which are often the hardest parts of producing ads consistently. That makes it easier to generate usable output without relying on video editing skills or external help.

For me, Topview fits squarely into the execution layer of design. It’s not a place to explore new visual ideas or experiment with creative direction. It’s a place to produce ads that look and feel correct for platforms like TikTok and Instagram, quickly and repeatedly.

I would use Topview when the priority is speed, consistency, and volume, especially for performance marketing or e-commerce use cases. It makes sense for individuals, small teams, or businesses that need to ship creatives regularly and don’t want the overhead of a full production workflow.

I wouldn’t use it for brand storytelling, long-form video, or highly distinctive visual work. The constraints are deliberate, and they show. But if the problem is producing platform-ready ads without friction, it solves that problem well.

In the Spotlight

Recommended listen: Creative Strategy and How AI Is Changing the Game for Growth Teams - with Reza Khadjavi, Motion CEO

This episode explores how AI is collapsing the gap between strategy and execution. Reza talks about why the most effective leaders are staying hands-on and using AI to compress work, rather than delegating everything away.

What stood out to me is how clearly this maps to what we’re seeing across tools today. Execution is becoming easier to automate. Clarity, taste, and decision-making are not. The advantage is shifting to people who can move between thinking and doing without friction.

“The fear with AI is staying stagnant. The opportunity is massively increasing your leverage.”
– Reza Khadjavi, ~0:22

This Week in AI

A quick roundup of stories shaping how AI and AI agents are evolving across industries

Google introduces A2UI, an open-source protocol that lets AI agents directly interact with user interfaces. Instead of relying only on APIs, agents can now navigate and operate software the way humans do, pointing toward more flexible, cross-tool automation.

YouTube faces a growing AI content saturation problem, with low-quality, auto-generated videos flooding the platform. The piece highlights how generative tools are shifting incentives toward volume over originality, and why platforms may need stronger signals to protect quality.

Meta places a major bet on Manus, a new AI initiative focused on building more capable, end-to-end agents. The move signals Meta’s ambition to compete not just on models, but on full agent systems that can reason, act, and execute across workflows.

AI Out of Office

AI Fynds

A curated mix of AI tools that make work more efficient and creativity more accessible.

AutoDraw → A lightweight AI tool that turns rough sketches into clean icons and illustrations in real time. Useful for quick visual thinking, early concepts, or when you need simple graphics without design overhead.

Huemint → An AI-powered colour palette generator that creates cohesive colour systems based on brand personality and use context. Useful for building UI and brand palettes that work well together without manual trial and error.

AI Text to Image Generator → An AI image generation tool focused on producing clean, usable visuals from text prompts. Practical for generating product images, backgrounds, or visual assets that can plug directly into design or marketing workflows.

Closing Notes

That’s it for this edition of AI Fyndings. With Motion structuring how work gets scheduled and followed through, Trickle turning ideas into product-shaped experiments, and Topview standardising how short-form ads are produced, this week highlighted a clear shift: AI is no longer just assisting work, it’s defining how execution happens.

Thanks for reading! See you next week with more tools, patterns, and ideas that show how AI continues to reshape how we plan, build, and create.

With love,
Elena Gracia
AI Marketer, Fynd

Some shifts in AI arrive with big product launches and louder promises. Others slip in quietly, reshaping habits we already take for granted. This week’s shift sits inside shopping, something most of us do routinely, often without much thought. It’s starting to move the work of choosing, comparing, and deciding away from people and into systems acting on their behalf.

This edition is a little different. Once a month, we pause the weekly roundups and focus on one defining shift in AI, unpacking it with a little more depth and clarity.

In Perspective: This week, we’re looking at agentic commerce and what happens when buying shifts from something we actively do to something we delegate. As checkout moves into chat interfaces and agents start narrowing choices for us, commerce becomes easier, quieter, and more controlled by systems than people.

It’s a fitting place to pause and reflect as the year comes to a close.

Shopping used to be a physical experience.

You went to stores. You walked through aisles. You touched things. You tried them on. You waited your turn. You compared options by moving from one rack to another. Sometimes you enjoy it. Sometimes you don’t. But you were present the entire time.

Choosing took effort, but it also felt real.

When shopping moved online, that experience changed.

Suddenly, everything was available on one screen. More options than any store could ever hold. You could shop from home, at any time, without travelling or standing in queues. On the surface, it felt easier.

But that abundance came with a trade-off.

With so many options, choosing started to take longer. You opened more tabs. You read more reviews. You went back and forth, unsure if the next option might be better. Delivery meant waiting. Returns meant more waiting. Trying something on often meant ordering first and dealing with the outcome later.

Online shopping removed physical effort, but it replaced it with mental effort.

So the industry spent the next decade trying to reduce that load.

Delivery got faster. Prime, same-day, next-day, scheduled slots. Returns got easier, at least on paper. Interfaces improved. Reviews became richer. Filters got more granular. Size charts, fit prediction, AR try-ons, and comparison tables tried to answer questions before you had to ask them. One-click checkout and saved payment details removed friction at the end of the journey.

What we were doing in physical stores was slowly recreated online, step by step, inside a single app.

The experience improved. Shopping became easier, faster, and more predictable.

But even after all of this, one thing stayed the same.

You still had to do the work of choosing.

You still had to explore options, compare trade-offs, and decide when something was “good enough.” Convenience helped, but it didn’t remove the responsibility of choice. It only made the process more manageable.

That is the point where AI enters the story.

What agentic commerce actually is

Agentic commerce is not “better recommendations.” It is not “chat-based shopping.” It is a shift in who performs the work of shopping.

Instead of you discovering, comparing, and narrowing down, an agent does it for you. You provide intent and constraints. The agent turns that into actions: search, evaluate, shortlist, and sometimes purchase.

The easiest way to see the shift is to look at what OpenAI, Shopify, and Stripe launched in late 2025.

OpenAI introduced Instant Checkout in ChatGPT and the Agentic Commerce Protocol (ACP) as “first steps toward agentic commerce.” Stripe announced it was powering Instant Checkout and co-developing ACP as an open standard to connect agents, people, and businesses for purchases.

Then Shopify made the point unmissable: Shopify merchants will be able to sell directly through ChatGPT conversations, no links or redirects.

This is the clearest “line in the sand” so far: commerce does not have to end on a storefront. It can end inside the agent interface.

And once the interface becomes the agent, the traditional shopping journey starts to collapse.

Where it stands now

Right now, agentic commerce is real, but uneven.

Most deployments sit in one of three buckets:

1. Discovery and advice inside an app

Amazon’s Rufus is a good example. It sits within Amazon’s app, answers shopping questions, compares products, and recommends options using Amazon’s catalog and web data. It’s not autonomous buying, but it makes conversation a primary layer of shopping.

2. Delegation for support and post-purchase tasks

Klarna’s AI assistant began with refunds, cancellations, and disputes and now handles a large share of customer interactions. That matters because returns and refunds are where trust in commerce is won or lost.

3. Early transactional agents that trigger platform conflict

As we discussed in one of our previous deep-dive editions on agentic browsing, the Amazon vs Perplexity conflict is one of the clearest early signals of what happens when agents start acting on a user’s behalf instead of just assisting them. What looked like a policy or security dispute was really about control. Who owns the journey when an agent decides what to click, what to skip, and what to buy.

That same tension shows up even more sharply in agentic commerce.

So yes, we are early.
But we are past the toy phase.

What gets better for people

Agentic commerce will feel like progress in the places where shopping is not identity, it is maintenance.

Groceries. Household basics. Replacement parts. The “I need it by Friday and I don’t care about browsing” category.

The benefits are straightforward:

It reduces the small tax of modern life. Fewer tabs. Less comparison. Less second-guessing.

It handles constraints better than humans do when we’re tired.
“Under $X.” “Delivers this week.” “Easy returns.” “Reliable seller.”

It improves continuity.
Agents can remember what you returned, what you complained about, what you re-ordered, and what you never want again.

This is why the most convincing agentic commerce examples are not “buy me a luxury handbag.” They’re “help me run my household.”

Amazon is already pushing this behaviour through voice too. The Verge reported Alexa Plus features that can monitor prices and even auto-purchase when a price drops below a user-set threshold. That is a clear example of “shopping becomes a background process.”

What gets worse for people

The cost is not money first. The cost is agency and variety.

Agentic commerce narrows the world before you see it. That sounds like a feature until you realise that browsing is how people discover taste, learn alternatives, and form preferences.

The biggest risk is not that agents make bad decisions. It’s that they make decisions that are “fine” so often that you stop noticing what you are not seeing.

At that point, you’re no longer choosing.

You’re approving.

Three problems show up fast as this becomes mainstream:

1. Opacity: If an agent recommends one option, why that one. Price? Margin? Reliability? Sponsored preference? Training bias? Or Partnership incentives? In a marketplace, ads are visible. With an agent, influence mostly stays invisible.

2. Mistakes with confidence: The more autonomous the agent, the more dangerous errors become. Wrong size is annoying. Wrong delivery address is a mess. The wrong seller can become a fraud problem. Commerce isn’t forgiving because it becomes physical.

3. Consent boundaries: Where does “help me find” end and “go ahead and buy” begin. That boundary needs guardrails, and right now those guardrails are inconsistent across products.

For example, buying clothes has always been about more than the item itself.

You might start with a clear need, but along the way you notice a different cut, a colour you hadn’t considered, a brand you didn’t know before. You compare how things look, not just how they perform. You change your mind halfway through.

An agent doesn’t do that.

If you ask for “a black dress for a wedding under $8,000 that is delivered this week,” you’ll get a sensible shortlist. But you won’t see the bright option that catches your eye. You won’t stumble onto a silhouette you didn’t know you liked. You’ll get something appropriate, not surprising.

The purchase will work.
But the experience will be thinner.

This is the quiet trade-off.

What it means for platforms

Agentic commerce attacks the foundation of large marketplace platforms, where ads, rankings, and merchandising shape what gets seen and bought.

Marketplaces like Amazon, DoorDash, and delivery aggregators don’t just earn money from transactions. They earn money from shaping choices before checkout. Ads, rankings, bundles, cross-sells, and sponsored placements exist because humans browse and linger.

Agents don’t linger.

When an agent becomes the interface, those surfaces lose power. The platform risks becoming an inventory backend rather than a decision shaper.

That’s why platform responses are predictable. Build their own agents. Restrict third-party agents. Redesign monetisation so agents, not humans, are the audience.

The Amazon–Perplexity fight wasn’t about one tool. It was about who owns the journey.

What this means for vendors and storefronts

So far, we’ve looked at agentic commerce from the buyer’s side. But the shift is just as meaningful on the other end of the transaction.

For vendors, agentic commerce quietly changes what a storefront is optimised for.

Traditionally, storefronts were built to influence humans. Layouts, imagery, copy, reviews, bundles, and nudges all existed to shape attention and guide choice. Browsing time mattered. Storytelling mattered. Even friction sometimes helped.

Agents don’t engage with storefronts that way.

When the buyer is a system, persuasion takes a back seat to clarity. What matters most is whether the product can be clearly understood, compared, and trusted without human interpretation. Clean attributes matter more than clever copy. Consistent pricing and delivery promises matter more than visual polish. Edge cases and vague policies become liabilities.

You can already see major storefront and payment platforms adjusting to this reality.

Shopify’s move to enable direct checkout inside ChatGPT is a clear signal. It’s not just about adding another sales channel. It’s about making Shopify-powered catalogs, pricing, and checkout callable by agents, without forcing the user through a traditional storefront. The store becomes the infrastructure, not the destination.

Stripe is approaching the same shift from the payments layer. With APIs designed for programmatic checkout and standards like the ACP, Stripe is positioning itself as the execution layer agents can rely on. The emphasis isn’t UI. It’s reliability, compliance, and safe automation.

PayPal and similar payment platforms are experimenting in a quieter way. Their focus is less on discovery and more on making authorization, identity, and trust portable across agent-driven flows.

And for commerce platforms like Fynd, this shift is even more structural. Unified catalogs, real-time inventory, order management, logistics, and policy enforcement aren’t just conveniences anymore. They’re prerequisites. Agents can only act when the underlying system reflects reality accurately. In that sense, Fynd and similar platforms are already closer to being agent-ready than many storefront-first tools.

This creates a real shift for teams building commerce products.

Instead of asking “how do we get users to explore more?”, the question becomes “can an agent reliably evaluate us?”

Can it understand our variants?

Can it predict fulfillment accurately?

Can it trust our return and refund logic?

For vendors with strong operations, this can be a positive change. Agents tend to reward reliability. Products that consistently deliver what they promise are more likely to surface, even without heavy brand spend. In that sense, agentic commerce can level parts of the playing field.

But there are risks too.

Agents optimize for predictability. That means newer brands, unconventional products, or anything that requires explanation can struggle to appear. Discovery narrows. Novelty gets filtered out early. Over time, this can quietly flatten marketplaces.

There’s also a loss of visibility.

When decisions happen inside agents, vendors may no longer know why they were shortlisted or excluded. A change in an agent’s logic, partnership, or ranking signal can impact sales without clear feedback. For storefront teams, debugging performance becomes harder when the “user” never actually sees the page.

The role of the storefront shifts as a result.

It becomes less about persuasion and more about being a source of truth. Less about telling a story, more about representing reality accurately enough for a system to decide. Storytelling doesn’t disappear, but it moves elsewhere. At the point of transaction, clarity beats charisma.

For vendors and commerce platforms, this is not a cosmetic change. It’s a structural one. The better your systems reflect how your business actually operates, the better agents can work with you.

What it means for brands and merchants

For merchants, this can be both a gift and a threat.

It’s a gift because agents can reward true performance: clean product data, fewer returns, reliable fulfilment, real availability. When the buyer is a system, sloppy operations become a liability faster.

It’s a threat because discovery becomes a gatekeeper problem. Instead of “how do I rank on Amazon,” it becomes “how do I get shortlisted by the agent.”

Shopify’s OpenAI partnership is exciting precisely because it gives merchants a new surface: selling directly in ChatGPT conversations. But it also makes the new competition obvious: you’re now competing inside an interface where the agent decides what’s shown first, and the user may never scroll beyond the shortlist.

What brands will do next

Brands won’t sit still. Agentic commerce forces changes not just in tactics, but in how brands choose vendors, platforms, and even how they think about their own stacks.

The first shift is structural.

Brands will invest heavily in machine-readable truth. Structured attributes, accurate inventory, delivery promises, return rules, compatibility data, and policy clarity stop being “backend hygiene” and start becoming growth levers. If an agent can’t reliably understand a product, it simply won’t surface it.

This alone changes how brands evaluate commerce platforms.

Instead of asking, “Does this storefront look good?” or “Does this tool help with conversions?”, brands will ask:

• Can this platform expose my catalog cleanly to agents?

• Can it guarantee real-time inventory and fulfilment signals?

• Can it execute orders, refunds, and changes programmatically without breaking?

Platforms that can’t support agent-readable catalogs, programmatic checkout, and reliable post-purchase flows will slowly fall out of favour, even if they perform well today for human shoppers.

The second shift is optimization for agents, not people.

Just as SEO reshaped content for search engines, we’ll see the rise of agent optimisation. Brands will start asking different questions. How often do we get shortlisted? What signals get us dropped. What does “reliable” mean to this agent?

This pushes brands toward platforms and vendors that offer observability. If decisions are happening upstream, brands will need tools that explain why they were chosen or excluded. Black-box systems become risky when agents, not users, are making the call.

The third shift is in how brands think about differentiation.

In an agentic flow, storytelling at the point of purchase matters less. Brands will move narrative and emotional differentiation to other surfaces: social, community, content, creators. The commerce layer itself becomes more utilitarian. Clean, predictable, and optimized for execution.

This also affects build vs buy decisions.

Brands will be less interested in stitching together brittle point solutions and more interested in unified stacks that reduce ambiguity. Fewer integrations. Clearer primitives. Systems that reflect reality accurately, not optimistically. Reliability beats flexibility when an agent is doing the choosing.

Then monetisation re-enters.

Paid placement won’t disappear. It will move upstream. Preferred suppliers. Trusted partners. Agent-friendly defaults that never look like ads on a screen. Brands that understand this early will start budgeting for “agent surfaces” the same way they once budgeted for search or marketplaces.

The shelf moves again.

This time, it moves into systems most customers never see.

And for brands, the real challenge won’t be learning how to sell to people. It will be learning how to be legible, trustworthy, and competitive in a world where the buyer is no longer human.

All of these shifts on the brand side depend on something deeper. They only work if the underlying commerce platform is designed to support agent-driven interactions in the first place.

Which brings us to the role of platforms.

Where this leaves platforms like Fynd

Agentic commerce doesn’t mean platforms like Fynd need to reinvent themselves. If anything, it validates the direction they were already moving in.

Agents only work when the underlying commerce stack is solid. Product data needs to be structured. Inventory needs to be real-time. Pricing, orders, shipping, refunds, and policies need to be connected. Online, offline, and marketplace channels need to operate as one system.

That’s already the foundation of unified commerce.

The opportunity for Fynd Commerce isn’t to build an agent. It’s to be the system agents plug into. Agents don’t replace commerce stacks. They depend on them.

What does change is the interaction layer.

Discovery, comparison, and even checkout are moving into conversational surfaces. Control and fulfilment still sit with the merchant stack. For Fynd, this means treating conversation as a new channel and agents as a new kind of marketplace, similar to Amazon integrations, but agent-native.

Practically, that means catalogs need to be agent-readable. Pricing, availability, and delivery promises need to be queryable. Checkout needs to be executable programmatically, with clear guardrails. Pages matter less. Primitives matter more.

The most meaningful agentic opportunity, though, isn’t on the consumer front end.

It’s in the back office.

Merchants don’t just need help selling. They need help deciding. What to restock. What to discount. Why conversions dropped. Which SKUs are causing issues. This is where agents can deliver immediate value, and it’s where Fynd already has context across catalog, inventory, orders, and operations.

Customer support is another obvious wedge. Not FAQ bots, but agents that can actually act. “Where is my order?” answered with real status. “Change delivery address” executed. “Initiate refund” triggered. This is where agentic commerce becomes operational, not cosmetic.

All of this points to a subtle positioning shift.

Unified commerce was about giving humans one system to manage complexity. Agentic commerce is about making that system operable by humans and agents together. That brings new requirements: permissions, guardrails, auditability, and clear points for human override.

The next narrative isn’t “AI features.”

It’s commerce that agents can safely operate.

In that sense, Fynd Commerce isn’t competing to be the agent. It’s becoming the infrastructure that makes agentic commerce possible.

Security, fraud, and “agent abuse”

Once agents can transact, commerce becomes a security problem in a new way.

The risks include:

• compromised agent accounts making purchases

• prompt-injection style manipulation through web content

• merchants being impersonated

• “agentic arbitrage” and exploit patterns where agents behave predictably

OpenAI and Stripe are clearly aware of this. ACP is positioned as a secure, merchant-controlled protocol where merchants can accept/decline orders and maintain compliance signals. That is a sign that the industry expects fraud and misuse to shop up quickly.

We’re also already seeing how quickly trust collapses when AI touches pricing and fairness. Instacart just ended its AI-driven pricing experiments after criticism and regulatory scrutiny around inconsistent prices shown to different users for the same items.

That wasn’t “agent checkout.” But it shows the same truth: automation in commerce creates trust failures faster than it creates delight.

When does this become “normal”

Two things can be true at once:

• we are early

• the timeline is not far

The most realistic adoption path looks like this:

Now to 12 months: agent-led discovery and assisted checkout expands across large surfaces (ChatGPT checkout, platform assistants, retailer copilots). We’re already here.

12 to 24 months: routine purchasing becomes increasingly delegated: replenishment, subscriptions, repeat carts, price monitoring, and “buy again” flows become agent-driven defaults (especially in voice and chat interfaces). Alexa auto-purchase is an early form of this.

24 to 48 months: more meaningful autonomy arrives, but with guardrails: approvals, spending limits, merchant accept/decline, auditable logs. B2B also accelerates because procurement is already structured and rule-driven. Forrester predicts agent-led quote negotiations will force a meaningful share of B2B sellers to engage.

I wouldn’t claim “complete autonomy everywhere” on a date. The limiting factor is not model capability. It’s trust, fraud, regulation, and who accepts liability.

My take

Agentic commerce will win where shopping is a chore.

But if we let the same logic dominate expressive categories, commerce becomes narrower and more homogenised. The “fun” of exploring is not just an activity or burden. It’s how people discover alternatives, develop taste, and avoid becoming trapped in defaults. Think about how often people find products on Instagram not because they searched for them, but because something caught their eye while scrolling.

The biggest danger is silent narrowing. Not bad outcomes, but repeated acceptable ones.

The most important design problem in agentic commerce is not checkout. It’s the moment where the agent needs to step back and let the human explore again.

If we get that right, agentic commerce becomes a powerful assistant.
If we get it wrong, it becomes a quiet gatekeeper.

In the Spotlight

Recommended watch: The Agentic Commerce Revolution is here - but it’s not what you think! (Brex CEO Pedro Franceschi)

I found this episode really interesting because it frames agentic commerce in the most practical way I’ve heard so far. Instead of treating it like “AI shopping,” Pedro explains it as orchestration. In B2B, buying isn’t one decision. It’s a chain of policies, approvals, procurement checks, security reviews, and accounting work that happens before and after the swipe. His point is simple: agents don’t just help you choose, they can absorb the messy coordination that makes corporate spend slow and painful. It also makes a strong case that the real opportunity is in B2B, where the complexity is higher and the ROI is easier to justify.

There’s 1.8 trillion corporate cards spent in the US today.
– Pedro Franceschi, Brex CEO • ~0:16

Closing Notes

That’s it for this edition of AI Fyndings. From shopping moving from exploration to delegation, to agents beginning to narrow choices before we ever see them, to platforms and brands adapting to a world where systems increasingly decide what counts as “best,” this week was about commerce quietly shifting out of our hands and into the logic of machines.

Thank you for reading, and happy holidays! See you next week with more stories, tools, and ideas shaping how AI continues to redefine how we build, choose, and live with technology.

With love,
Elena Gracia
AI Marketer, Fynd

Welcome to AI Fyndings!

With AI, every decision is a trade-off: speed or quality, scale or control, creativity or consistency. AI Fyndings discusses what those choices mean for business, product, and design.

In Business, Radix shows how brand visibility now lives inside AI-generated answers rather than on traditional search pages.

In Product, Marengo 3.0 makes video searchable and structured, turning scattered footage into something you can actually work with.

In Design, Wan offers controlled, high-fidelity visuals that help teams explore ideas without losing consistency.

AI in Business

Radix AI: Measuring Brand Visibility Inside AI Answers

TL;DR

Radix shows whether your brand appears inside AI-generated answers. It is an early AEO signal that helps you understand if AI tools include you in the conversation.

Basic details

• Pricing: Free plan available

• Best for: Marketing teams, founders, product teams

• Use cases: AEO visibility, competitive benchmarks, topic-level presence

AI has changed how people search. Instead of scanning links, users ask ChatGPT, Perplexity or Google AI a question and get one consolidated answer. That answer becomes the full experience and it is not obvious which brands shaped it.

This is where AEO becomes relevant.

SEO helps you appear on a page.
AEO helps you appear within the answer.

Radix is built around this shift. It tracks how often your brand appears inside AI responses and the topics where those mentions happen. Instead of measuring position, it measures presence. This is becoming the new baseline for visibility as LLM-led search grows.

What’s interesting

Radix surfaces a layer of visibility that most teams do not measure today. It shows whether AI tools mention your brand, how often that happens, how you compare with competitors and which topics you are most associated with. When I looked at the dashboard, I could see how often a brand appeared across ChatGPT, Perplexity, and similar tools, how it ranked within its category, and how its share of voice compared with competitors. It reframes visibility from “what page am I on?” to “do I even show up in the answer?”

This reframes the idea of search visibility. It is no longer about ranking on a page. It is about whether you are part of the answer at all.

The topic-level breakdown is also useful. It shows which themes your brand is most visible in, such as inventory management or retail technology. This helps differentiate between areas where you genuinely appear and those where you have little or no visibility.

Another really helpful layer is industry ranking. Radix would take up your brand and position it amongst your competitors ranking for the same keyword, making it easier to map which companies make it to the most searches and who barely appears.

Across these features, Radix feels designed around a simple question: Is your brand part of the AI conversation or not? The framing is new, but it reflects how search behaviour is actually shifting.

Where it works well

Radix works for teams who want to understand how their content performs in generative search. This includes marketing teams, growth teams, founders and SEO teams who want an early read on AEO behaviour.

It is especially useful in industries where users research before purchasing. SaaS, finance, logistics, education and similar categories fall into this group. Users here often rely on AI tools to explore options or compare solutions.

The trend view helps teams understand whether visibility is improving or declining. The competitor view grounds these shifts in context.

Overall, Radix is strongest as an AEO baseline. It tells you if you show up, how often and in which contexts.

Where it falls short

Radix shows when you appeared. It does not explain why.

It does not reveal which queries triggered your mentions or how you were positioned. A mention may be an example, a comparison or a recommendation, but Radix treats all of these equally. This can limit how strategic the insights feel.

Radix also does not break visibility down by intent. Informational queries behave differently from category queries or troubleshooting questions, but the platform combines them into one view.

And while it shows what is happening, it does not offer guidance on how to influence the outcome. Teams still need to interpret the patterns on their own.

What makes it different

Radix is intentionally narrow. It measures AEO visibility and nothing beyond that. It does not analyze prompts, interpret the model’s reasoning or offer optimization advice.

Profound is the tool most similar to Radix in this space. Profound is a diagnostic AEO tool that reveals the prompts and reasoning behind a brand’s mention inside AI answers.

This difference becomes clearer in practice.

Profound explains why a model mentioned a brand. It shows which prompts triggered the mention, how the model interpreted your content and how you were positioned relative to others.

Radix shows if you were mentioned. It focuses on frequency, presence and category share, not reasoning.

Profound helps you understand the logic.
Radix helps you see the signal.

Radix is simpler and easier to adopt, but also more limited. It gives you the what, not the why or the how.

My take

Radix solves a new and necessary problem. It shows whether AI models surface your brand inside the answers they produce. The interface is clean, the comparisons are helpful and the framing matches how search is evolving.

The tool still requires human interpretation. Radix does not explain how you were described or why visibility shifts over time. And because AI-generated answers change often, the signal can feel unstable.

Even so, Radix fills a gap that most teams are not addressing. It does not replace SEO or analytics, but it adds a visibility layer that becomes more important as AI-generated answers become the first place users look.

AI in Product

Marengo 3.0: Video Search That Finally Feels Practical

TL;DR

Marengo makes video searchable and easier to understand. It helps you find key moments across large sets of clips without watching everything manually.

Basic details

• Pricing: Free trial available

• Best for: Teams handing large or repetitive video collections

• Use cases: Video search, dataset exploration, research review, lesson or training analysis

Video has quietly become the default way teams document what they do, from product demos and research recordings to customer calls and internal walkthroughs. But despite how common it is, video remains the least usable format in a workflow. You can skim a document, search a Slack thread, or filter through a dataset, but a video still demands your full attention. You cannot query it, you cannot glance through it, and you rarely know where the information you need actually sits.

Marengo 3.0 is built to improve that experience. It breaks a video into meaningful pieces, understands what is happening in each scene, and lets you surface those moments using natural language instead of a timestamp scrubber. It can summarize clips, find similar scenes, and map how different videos relate to one another.

What’s interesting

Marengo handles simple prompts surprisingly well. When I typed a short description of the Hogwarts courtyard, it located the correct moment without much trial and error.

The upload and indexing setup added to this. Each file has to meet specific requirements, and once uploaded, I could group multiple videos into an index. This immediately shifted the experience from handling individual clips to working with a defined dataset, which made the tool feel more like an analysis environment than a basic player.

The Embeddings view was another standout. At first the scatter plot looked random, but patterns started to appear once I interacted with it. Similar scenes grouped together, unrelated ones spread apart.

The Analyze feature also stood out, though in a different way. It captured the flow of a scene well enough to avoid rewatching everything, even though some character details were incorrect.

Across these features, Marengo shows structure inside video that you usually cannot see by scrubbing.

Where it works well

Marengo works best when exploring multiple videos inside an index. I could describe a moment and the tool searched across the entire set at once, making it easier to handle raw footage and long recordings.

The embedding view gives a quick overview of variety and repetition across videos. Once you understand the layout, it becomes an effective way to see similarities before digging deeper.

This makes Marengo useful for:

• raw shoot review

• research footage

• lesson recordings

• user-generated content

It seems best suited for people who work with video in a functional way rather than a creative one. Product teams, researchers, educators, and anyone handling documentation or reference footage would benefit from being able to search across a set of clips and get an early sense of what the material contains. It is not a tool for finishing touches or production work, but it is helpful in the early stages when you are trying to understand what you have.

Where it falls short

Jumping between pictures wasn’t really smooth. After running a search, I could not directly continue into analysis. I had to manually pick the video again.

Summaries capture the flow of a scene but cannot be trusted for specific details. In the Hogwarts clip, the tool misidentified Madam Hooch as Professor McGonagall.

The embeddings view took time to understand. Some unrelated clips appeared close together, and the interface did not offer deeper cluster controls.

Another practical limitation is that Marengo does not accept URLs. Every video must be downloaded and uploaded manually.

These issues make the workflow feel heavier than necessary.

What makes it different

Marengo focuses on understanding existing video, not creating new footage. Tools like Runway specialise in generation and editing, while Marengo focuses on scene-level analysis and retrieval.

It also works at a finer level of detail. Instead of treating each video as one unit, it breaks it into smaller clips that can be searched and compared.

Overall, Marengo treats video as information, not content, which places it in a very different category from most AI video tools.

My take

Marengo 3.0 is a meaningful step forward, especially with indexing and cross-video search. It makes large collections easier to navigate and handles straightforward prompts reliably. But the product around the model still feels early. The features do not connect smoothly, summaries miss details, and the embed view needs more clarity and control.

Even with these gaps, the direction is right. Marengo is trying to solve a real problem: making video easier to understand. It is not something I would rely on for precise information yet, but it does make exploration faster and more structured. With more refinement, it could become a strong companion for teams who handle video at scale.

AI in Design

Wan: High-Fidelity Video Generation for Designers and Storytellers

TL;DR

Wan creates stable, high-quality images and short videos. It is best for exploring early visual ideas before moving into heavier production tools.

Basic details

• Pricing: Free tier available; paid plans vary based on credit usage.

• Best for: Designers, content teams, creative strategists

• Use cases: Product mood boards, short concept videos, image-to-video, character shots, voice-driven scenes

Most teams experimenting with AI video run into the same problem. The tools rely heavily on open-ended prompts, which produce results that are visually impressive but rarely consistent ones.

Camera motion shifts unpredictably, lighting changes between runs, and characters drift out of frame. This makes it difficult to use AI video in real design or content workflows where reliability matters.

Wan tries to solve this by giving structure to the entire process. Instead of leaving everything to the model, the tool offers fixed camera movements, lighting presets and scene styles that act as boundaries. You decide how the shot behaves and the model fills in the details. This reduces variation and makes the output easier to iterate on.

Wan is a browser-based generator for images and short videos. It focuses on producing stable, high-quality visuals quickly. You can create a still image, turn it into a short clip, or drive a character with audio without switching tools. It is not built for full production, but it is effective for teams who need dependable visuals during planning, exploration or creative reviews.

What’s interesting

Wan is interesting because it makes controls that usually feel complex much easier to use.

The interface shows you exactly which camera moves, lighting setups and aspect ratios are available. This clarity helps reduce trial and error, especially when teams are exploring several directions at once.

When I tested a 5-second AI video generation, Wan automatically added audio even though I did not provide a script. This was unexpected, but it showed how the tool fills in certain creative elements on its own.

You can generate a still frame, turn it into a short video and quickly test different camera paths or styles without starting over. This makes Wan helpful during creative reviews where multiple versions of an idea need to be evaluated side by side.

Another useful feature is the Avatar, which offers 3 functions: Speech-to-video, Character swap, and Animate swap.

• Speech-to-Video lets you upload an image of any character, person or persona and animate it using an audio file. The tool applies facial motion, timing and expression to the static image, and it lets you choose from different accents to match the voice. The Mona Lisa example makes this clear. A single still image becomes a short, expressive clip that mirrors the tone of the audio.

• Character Swap replaces the subject in a scene while keeping the original shot structure.

• Animate Swap applies the movement from one clip to another subject, which helps when testing gestures or pacing.

These tools don’t replace animation software, but they make early idea testing much easier.

Where it works well

Wan works well when you want to explore ideas with clarity rather than speed. Even though rendering takes time, the tool’s structure makes each version predictable. This helps when teams need to refine tone or understand how a moment might look before committing to a final approach.

It is especially useful for short narrative moments. Simple gestures, small reactions and basic scene ideas come through with enough detail to support discussions in design, storytelling and content planning. The Avatar tools extend this by letting you animate a still image, try different voices or swap characters without rebuilding the scene.

Wan also fits into workflows where comparison matters. Changing the camera path or adjusting the lighting produces variations that hold their structure, which makes side-by-side review easier. For early creative alignment, this reliability becomes more valuable than speed.

Where it falls short

Wan can only create clips up to fifteen seconds, which keeps it useful for short moments but not for longer scenes. If you need continuity or multiple beats in a story, you will still need other tools to put everything together.

Like most AI video generators, Wan also depends on precise prompts. You have to be clear about what you want. Even then, the output may be close but not exact, so there is still some trial and error involved.

The tool is slow. It shows a countdown, but a single video can take more than five minutes to finish. This makes it harder to test many versions quickly, especially during fast-moving projects.

Another limitation is accuracy. Simple movements work well, but complex actions, hand interactions or fast motion can create distortions.

What makes it different

Wan is different because it makes AI video generation feel predictable. Most generators either give you too much freedom, which leads to random results, or they give you very strict controls that are hard to use. Wan sits in the middle. It gives you enough structure to keep shots consistent, but it is still simple for any designer or content creator to pick up.

This makes Wan useful at the start of a creative process. You can test an idea, change the camera or lighting, and see new versions that actually relate to each other. Tools like Runway or Kling often introduce big visual changes from one run to the next, which can make early exploration harder.

Wan’s limits become more profound once you move beyond early ideas. The clips are short, detailed motion is not always accurate and the output is not polished enough for final use. That is when teams usually shift to tools like Runway for stylised videos or Kling for more realistic shots.

My take

Wan is a useful tool for the early stages of creative work. It does not try to replace production software, and it should not be used that way. Instead, it gives teams a reliable way to explore ideas before they become fully developed. The structure of the tool helps keep outputs consistent, and the quality is strong enough to guide discussions about tone, story and direction.

The limitations are real. It is slow, the clips are short and the accuracy falls off when the motion gets complicated. But these issues do not take away from its value. Wan works best when you need clarity, not polish. It helps you see how a moment might feel, how a character might behave or how a scene could be framed, long before you invest in a heavier workflow.

For teams who think visually, this kind of early insight matters. Wan gives you a starting point you can react to, refine and build on. And in a space where most tools still feel unpredictable, that reliability is genuinely helpful.

In the Spotlight

Recommended watch: How to Dominate AI Search Results in 2025 (ChatGPT, AI Overviews & More)

This video offers an insight into AI search that feels grounded rather than hype-driven. Matt Kenyon breaks down how generative tools actually pull information, why top-ranking pages still shape AI answers, and how structure and schema make your content easier for models to use.

What stood out to me is how familiar the playbook is. AI search doesn’t replace good SEO but rather reinforces it. The fundamentals matter even more now.

Ranking in AI search pretty much boils down to just doing good SEO.
– Matt Kenyon, ~0:40

This Week in AI

A quick roundup of stories shaping how AI and AI agents are evolving across industries

Google releases an updated Gemini Deep Research, a more capable agent that plans, searches, and synthesises information into detailed reports. It signals Google’s push toward deeper, autonomous research workflows.

OpenAI introduces GPT-5.2, a long-context model built for coding, agents, and knowledge work. It focuses on stronger reasoning and more reliable multi-step execution across complex tasks.

xAI launches the Grok Voice Agent API, enabling real-time speech interaction for Grok-powered assistants. Developers can now build voice-first agents with fast responses and flexible voice options.

AI Out of Office

AI Fynds

A curated mix of AI tools that make work more efficient and creativity more accessible.

Diagramming AI → A lightweight tool that turns concepts, flows, and structures into clean diagrams with minimal effort. Useful for mapping ideas before they turn into full designs.

ColorHub → An AI-powered palette tester that checks contrast, accessibility, and performance across themes and modes. Helpful for stress-testing colors inside a design system.

Christmas Background Generator → A simple browser tool that adds festive, high-quality Christmas backgrounds to photos. Great for quick seasonal visuals without manual editing.

Closing Notes

That’s it for this edition of AI Fyndings. With Radix reframing how brands show up inside AI answers, Marengo making video searchable instead of overwhelming, and Wan giving creative teams more control over early visuals, this week pointed to a simple shift: AI is helping us see what we couldn’t see before.

Thanks for reading! See you next week with more tools, shifts and ideas that show how AI continues to reshape how we understand, create and decide.

With love,
Elena Gracia
AI Marketer, Fynd

Welcome to AI Fyndings!

With AI, every decision is a trade-off: speed or quality, scale or control, creativity or consistency. AI Fyndings discusses what those choices mean for business, product, and design.

In Business, Cluely rethinks how real-time support shows up inside meetings.
In Product, Visily brings structure to early ideas before design even begins.
In Design, Rive shows how motion can respond to users instead of playing on repeat.

AI in Business

TL;DR

Cluely is a live meeting assistant that listens, captures your screen, and notes conversations as they happen. What I found useful is that it works well for anyone who needs to take back-to-back calls or even students who need quick clarity without multi-tasking.

Basic details

• Pricing: Free plan available; Pro plans start at $20 per month

• Availability: Browser-based

• Best for: Students, sales teams, professionals

• Use cases: Meeting notes, summaries, and follow-up emails

Most meeting tools help you only after the call ends. Cluely comes from a very different starting point. Its earliest version was built by students who noticed how difficult it is to think, listen and take notes all at once during high-pressure situations like job interviews. That early experiment caused controversy, but it led to a simple insight: people often need help during the meeting, not after they’ve already forgotten half of it.

Cluely is built around that idea. Instead of being a passive recorder, it tries to support you while the meeting is happening. It listens, watches your screen and tries to keep up with the mess in real time. It exists for the exact moments you won’t admit out loud, like when you are presenting and missing half of what people say, or when you zone out for thirty seconds and return to a discussion that has somehow moved ahead four chapters without you. Cluely helps you circle back to the plot without asking you to confess that you lost midway.

It also cuts out the emotional work of pretending you can keep up with everything. Normally you are multitasking your way into a headache. Cluely takes some of that pressure off so the meeting or lesson feel less like a survival test.

What’s interesting

Cluely’s most interesting quality is how active it is while you are in the call. You see a running transcript in real time, with a claimed 300 millisecond response time. It supports more than 12 languages, including English, Chinese and Spanish, which is great for global teams and also for that one coworker or speaker who switches languages mid sentence like they are changing TV channels.

Alongside the transcript, Cluely lets you ask questions at any moment such as “What should I say?”, “What’s the summary so far?” or “Give me follow-up questions.” You can even trigger help instantly with keyboard shortcuts, which feels natural once you get used to it.

The tool also picks up on whatever you’re showing on screen. Slides, documents, resumes, scripts or briefs you upload into a mode become reference points during the call. As you talk, Cluely helps surface what’s relevant so you don’t break focus.

Once the meeting ends, everything comes together quickly. You get a summary, action items, suggested next steps, a drafted follow-up email and a complete transcript you can chat with for clarity, and all of this can flow into tools you already use, like Slack, HubSpot, Salesforce or Notion. The turnaround time makes it feel like a companion that stays one step ahead.

Where it works well

Cluely works best in meetings that move quickly or jump between different topics. If you’re presenting, switching tabs or walking someone through a document, it keeps track of the important parts so you don’t have to. It’s especially helpful in everyday internal calls where you just want the key points captured without worrying about accuracy.

The different modes make it easier to use in specific situations. Sales Mode brings up helpful prompts during customer calls. Recruiting Mode picks out useful details in interviews. Lecture Mode focuses on the main ideas instead of tasks. Each mode gently nudges the tool to match the situation, something most meeting assistants fo not even attempt.

The meeting brief you get afterward is immediate and digestable. It shows decisions, action items and next steps right away, so you can move into your next call without trying to remember what happened. Everything is already laid out for you.

Where it falls short

Cluely is helpful, but it is not a mind reader. When a conversation moves very fast or gets complicated, its suggestions can feel a bit off or arrive a little late. It usually captures the main idea, but not the finer details, especially in technical discussions or any meeting with a lot of context.

The notes it creates are useful, but they are not ready to share as-is. You still need to read through them and make small corrections.

And although the tool supports multiple languages, multilingual accuracy depends on how clearly participants speak and how often they switch languages mid-sentence. Speaker separation can also be confusing in group calls where people talk at the same time or audio quality varies.

It also may not be the best choice for sensitive meetings. Even though you can control when it listens or records, some teams prefer to keep tools like this out of client calls, private conversations or situations with strict rules around confidentiality.

In short, Cluely works well when the goal is quick clarity, but it is not built for technical, fast-moving or sensitive conversations.

What makes it different

Cluely’s edge is that it is genuinely helpful during the call, not only after. You can ask questions, request a quick recap or get clarification in the middle of the conversation, which makes the tool feel genuinely useful in real time.

It also handles small but important parts of a meeting. It captures follow-ups, drafts emails, organises action items and lets you explore the transcript once the call is over. All of this can move into the tools you already use, like Slack, HubSpot, Salesforce, or Notion, which means you’re not juggling five different systems after every call.

Cluely is not built to run your entire workflow. It aims to make meetings easier to keep up with and less mentally tiring.

My take

Cluely solves a familiar problem, the gap between what was said and what you actually remembered. It helps in the moments where your brain simply cannot keep up. When you are presenting, your attention is split across screens, reactions and the fear of sharing the wrong tab, so you miss half the conversation without meaning to. And on the other side, there are the quieter moments where you drift for a few seconds and suddenly realise the discussion has moved three layers deeper. Cluely gives you the thread back without making you announce that you lost it.

Outside this, the tool is helpful but not something extraordinary. It will not decode complicated discussions or turn a messy meeting into a good one. But when you use it in the moments where most people actually struggle, it keeps you sane. It understands the call, remembers the important points and helps you move to the next task without carrying the entire conversation in your head.

AI in Product

Visily: UI Design for the Rest of Us

TL;DR

Visily turns prompts, screenshots, and sketches into editable UI screens, helping non-designers and fast-moving teams get to their first-workable draft quickly. It stands out for generating clean structures early-on, without overstepping into creative decisions.

Basic details

• Pricing: Free plan available; Pro starts at $14 per editor per month

• Availability: Browser-based

• Best for: Non-designers, product managers and founders

• Use cases: Quick UI drafts, recreating reference screens and mapping product flows

There’s a specific moment in product building that everyone hates: when you know what you want, but you have nothing to show for it. You say, “Imagine a simple grocery app home screen,” and someone nods, but no one really knows what you mean. Visily is built for exactly that moment.

At its core, Visily is an AI-assisted UI design tool that helps you create screens quickly without needing design experience. You describe the screen, upload a competitor’s screenshot, or sketch something on paper, and the tool converts that into an editable layout. The objective is simple: move ideas from vague to visual without slowing down the team.

What’s interesting

The most interesting part of Visily is how practical it feels. I tested it using the kind of prompts you normally say out loud in a meeting. A simple line like “grocery app home screen with product categories” generated a layout that looked like a baseline wireframe.

It gave me something clear enough to react to without needing to overexplain.

The screenshot feature was even more useful. I uploaded a screen from an existing app and wanted to see how well Visily could break it down. It did not simply trace the layout. It separated the screen into editable pieces such as headers, cards, buttons and text blocks. This cuts out a lot of manual effort, especially during competitive research.

Another detail that stands out is how Visily handles smaller tasks that usually slow you down. It renames layers, adjusts spacing, creates dark mode versions and generates themes from text or images. These are not the features that draw attention, but they help maintain momentum.

Visily also stays within reasonable boundaries. It does not try to guess your entire flow or invent patterns that do not exist. Its output is stable and predictable, which helps build trust in what you are working with.

Where it works well

Visily works best at the beginning of the design process. This is the stage before branding decisions, high-fidelity work or long conversations about spacing. It helps teams turn early ideas into something they can see and discuss without waiting for a designer.

It is well suited for

• PMs and founders who want to visualise ideas quickly

• Engineers who need clarity on flow and layout

• Designers who want to skip repetitive setup and explore more directions

• Teams building internal tools where clarity matters more than finesse

It is especially helpful in fast brainstorming sessions. Instead of debating ideas verbally, you can generate a few variations in minutes and immediately see what direction works. It keeps the focus on understanding the idea rather than the mechanics of drawing it.

Where it feels short

Visily works well when the screens are simple, but it struggles when layouts become more detailed or unusual. It handles common mobile and web patterns without issues, but dashboards, multi-column pages, and admin-style interfaces do not come through as accurately. These screens usually need manual fixes because the tool streamlines the layout instead of capturing the full structure.

It can also get tedious to keep scenes consistent across a large project. When you restyle or regenerate a screen, small things like spacing, alignment, or text styles may shift. Visily has components, but they do not have the same flexibility as those in Figma. This makes it harder to reuse patterns or maintain a design within the tool.

Visily is ideal for smaller projects. As soon as you stack multiple screens on one board or have several people working together, the canvas can feel slower. This may seem common for browser tools, but it is still noticeable when you are moving quickly.

You can export work to Figma, but it is not a perfect transfer. Layers do not always import cleanly, and constraints often need reorganising. Designers usually spend a little time tidying up the file before they continue.

Since Visily is fully cloud-based, a stable internet connection is important. If the connection is weak, features like generation and editing slow down or pause.

What makes it different

The difference with Visily is much easier to understand when you compare how long the same tasks take in other tools. In Figma, recreating a reference screen can take 20-30 minutes of manual work; in Visily, you can get an editable version in seconds. Uizard offers something similar, but Visily’s output tends to preserve hierarchy and spacing more accurately, especially for mobile layouts.

The prompt-to-screen feature is also more controlled than most alternatives. Instead of generating an entirely new style each time, Visily keeps to a predictable structure, which makes it easier to iterate without starting over. This is useful when you’re exploring variations rather than looking for an entire redesign.

Another area where it stands out is how it handles smaller details automatically. Text styles, spacing, and basic grouping are applied consistently across generated screens. These aren’t some out of the box features, but they cut out routine steps that usually slow down early drafting.

And compared with tools like Uizard or Canva, Visily is clearly oriented toward product surfaces rather than marketing layouts. The components it produces such as cards, navigation bars, headers, form elements, match what you’d expect in real apps, not generic templates.

My take

Visily is a useful companion for teams who move quickly and don’t want to get stuck in the “I’ll mock this later” loop. It helps you get visual clarity early without pulling a designer into every initial discussion.

For early drafts, internal tools, flow mapping, and quick product conversations, it’s absolutely worth having in your toolkit.

AI in Design

Rive: Bringing Interactive Motion Into Everyday Product Design

TL;DR

Rive lets you create lightweight, interactive animations that react to user input, making it valuable for teams who want their UI to feel more responsive. It works best for small interactions that make a product feel clearer and easier to use.

Basic details

• Pricing: Free plan available; paid plans start at $9 per seat per month (required to ship your builds)

• Availability: Browser-based

• Best for: Motion designers, product designers and front-end teams

• Use cases: Micro-interactions, UI animations and responsive icons

I am not a motion designer, so my first impression of Rive was that it looked like another animation tool. But after speaking with a motion designer and trying it myself, I realised it solves a very different problem. Most animation tools create files that play the same way every time. Rive lets you build animations that respond to what a user does. That small shift changes how motion fits into a product.

For example, a button does not just animate once and stop. It can react when you hover, when you click or when you drag. An icon can change its state based on input. A loader can move differently depending on real progress. Instead of creating a single animation sequence, you create behaviour that feels connected to the interaction.

Rive is designed to help you build motion that feels smooth, lightweight and integrated into the product rather than added on top. Even as someone outside the design world, I could see why teams use it to bring more personality and clarity into their interfaces.

What’s interesting

The most interesting thing about Rive is how it blends design and behaviour. You are not just drawing frames. You are defining how the animation should respond. This can be done using The State Machine which lets you connect different animation states to triggers like hover, tap or drag.

What’s more is how smooth and lightweight everything feels. Rive animations run in real time across web, iOS, Android, Flutter and even game engines. You’re not exporting videos or large files. You’re exporting instructions for how motion should behave.

The editor also keeps design and animation in one place. You move between Design Mode and Animate Mode without switching tools, which helps keep the workflow consistent. What you create in Rive is exactly what developers implement, because you share the animation as a small embed snippet or code.

And when you look at Rive’s example library, you understand what the tool is built for: buttons that bounce, toggles that glide, icons that react when dragged, characters that move with input. These small pieces show how Rive changes the feel of a product without changing its structure.

Where it works well

Rive is most helpful in the parts of a product where small interactions make a big difference. This is usually when a simple visual response can guide the user, reduce confusion or make the interface feel more efficient.

It works especially well for things like buttons, icons, toggles and loading states because these elements need to react quickly and clearly. A subtle change in motion can tell the user that something was clicked, something is in progress or something has updated.

It is also strong in onboarding or educational flows, where movement can help direct attention without overwhelming the screen. Because Rive supports data-driven motion, animations can respond to live information too, which makes it useful in dashboards or metrics-heavy interfaces where visuals need to update smoothly.

In short, Rive works best in small, frequent interactions across a product. These are the places where responsive motion improves the experience without getting in the way.

Where it falls short

Rive is impressive, but it is not the easiest tool to pick up if you are new to motion. The idea of connecting different states and setting up how an animation should react takes some time to understand. If your background is in static design tools, the learning curve can feel noticeable at first.

It is also not the right choice for long or detailed animations. Tools like After Effects are still better for storytelling, complex sequences or anything that needs heavy visual effects. Rive works well with short, interactive moments, not in full cinematic motion.

There is a practical consideration too. Developers need to know how to implement Rive files in the product. The workflow is smoother than handing off video files, but it still requires coordination between design and engineering.

What makes it different

Rive stands out because it treats motion as part of the product experience rather than a final layer added at the end. Instead of designing an animation somewhere else and hoping it works once it is implemented, Rive lets teams experiment, test and refine motion earlier in the process. This makes animation feel more integrated and less like something that gets patched in later.

It also sits in a very specific place among its competitors.

Lottie is great for simple animations, but what you export is fixed. It plays, loops or reverses, but it cannot respond to user input.

After Effects gives you creative freedom, but the output is heavy and not suited for real-time product interactions.

Figma’s animation tools work well for prototypes, but they are not designed for production-level motion or long-term use in an app.

Rive fills the gap between these tools. Designers can define how motion behaves without coding, and developers can use the animation directly without rewriting it. This reduces the usual friction between design and engineering and makes motion feel like a natural part of the product, not an added effect.

My take

Rive is one of those tools that makes more sense once you see it in action. Even as someone who is not a motion designer, I could understand how much it helps with everyday product interactions. It makes buttons, icons and small UI moments feel clearer and more responsive, which is something most teams struggle to get right.

At the same time, it is not a plug-and-play tool. You need to spend time learning how it works, and not every team has the bandwidth or the motion mindset to adopt it smoothly. When the skills and use cases line up, the results are impressive. When they do not, the tool can feel heavier than expected.

For teams that care about the details of how a product behaves, Rive offers real value. For teams that only need simple visuals or basic animations, the extra effort may not be worth it.

In the Spotlight

Recommended watch: Why Cluely is leaning into real-time, context-aware AI

This podcast breaks down Cluely’s core idea in a way product demos don’t: an assistant that sees your screen, hears your calls and helps without waiting for a typed prompt. Roy Lee talks about why this feels like the next step beyond chatbots, how real-time context changes the value of AI and why Cluely is positioning itself as the application layer built to ride future models rather than compete with them.

​​The problem with ChatGPT right now is that you have this super intelligent model, but it can only answer the questions that you like to type in… why limit it to only the text question that you ask it?

– Roy Lee, CEO, Cluely ~0:51

This Week in AI

A quick roundup of stories shaping how AI and AI agents are evolving across industries:

Claude Code comes to Slack, giving teams the ability to generate, fix, and reason through code directly inside their day-to-day communication tool. It is a bigger shift than it appears, because it moves AI from “a place you go” to “a layer inside where you already work,” especially for engineering teams.

Google introduces Doppl, a new AI-powered discovery feed that pulls together outfit ideas, virtual try-ons, and shoppable looks in one place. Instead of searching for products manually, users can browse AI-curated styles that update as they interact, blurring the line between inspiration and commerce.

Workspace Studio brings AI agents into Gmail, Drive, and Chat, turning Google Workspace into an orchestrated system rather than individual apps. These agents can draft messages, organise files, suggest follow-ups, and help teams stay on track across tools, signalling Google’s push toward a more automated, assistant-driven workflow.

AI Out of Office

AI Fynds

A curated mix of AI tools that make work more efficient and creativity more accessible.

Visual Electric → A fast image-generation tool for quickly exploring visual styles and concepts.

Brushless → An AI sketching tool that turns rough drawings into clean, stylised illustrations.

AI Image Editor → A browser-based editor for instant background removal, cleanup and image enhancements.

Closing Notes

That’s it for this edition of AI Fyndings. With Cluely easing the pressure of fast-moving meetings, Visily helping ideas take shape faster and Rive bringing interaction to life, this week was all about improving the parts of work we often overlook.

Thanks for reading! See you next week with more tools, shifts and ideas that show how AI is slowly reshaping the way we work.

With love,
Elena Gracia
AI Marketer, Fynd

Welcome to AI Fyndings!

With AI, every decision is a trade-off: speed or quality, scale or control, creativity or consistency. AI Fyndings discusses what those choices mean for business, product, and design.

In Business, HoneyBook shows how AI can help small businesses stay on top of their everyday work without extra effort.

In Product, Bubble and Lovable reveal two different interpretations of what an AI builder should be. One careful and flexible, the other confident and opinionated.

In Design, Marble pushes past images and into full spatial environments, giving you a sense of a place rather than a frame.

AI in Business

HoneyBook: A System For The Space Between Tasks

Running a service business never feels as simple as it looks from the outside. People imagine a clean sequence where a client reaches out, you agree on the work, and everything moves along neatly. But the real effort lives in the spaces between those steps. A half-written proposal waiting for the right moment. A call you meant to schedule. A contract that should have gone out yesterday. A payment reminder that keeps slipping to the bottom of the list. None of it is overwhelming on its own, yet together they create a constant pull that follows you through the week.

HoneyBook is built for freelancers, independent professionals, and small teams. These are the people who feel that pull the most, because they manage every part of the client journey themselves. The tool is currently available only in the United States and Canada.

When I tried HoneyBook, the first thing that stood out was how it eased me into the setup. It didn’t burden me with menus or features. It simply asked what my business is about, how it runs, and what I want to fix first. Based on that, it built a starting point that felt familiar. It was practical and direct, and it made the rest of the platform easier to engage with.

What’s interesting

HoneyBook takes a process that is usually scattered across emails, spreadsheets, DMs, and calendars and turns it into one steady flow. When a new inquiry comes in, you don’t lose it under everything else happening that day. You open it and instantly see where it came from, what the person needs, and what should happen next. It is one of the few tools where the workflow feels followable even on a busy day.

The AI plays a quiet, supportive role. It does not try to take over your work or predict your decisions. Instead, it helps at the moments where people often slow down. A draft when you cannot find the right words. A reminder when a lead starts to fade. A quick outline before a call so you walk in with clarity. None of it feels like automation happening in the background. It feels like the small support you need to keep moving.

Where it works well

HoneyBook works especially well for service businesses that follow a steady process. Someone reaches out, you talk, you share details, they sign, you begin the work. The platform supports this flow without asking you to redesign it.

The lead forms are flexible and easy to create. Once a lead enters the system, HoneyBook organizes it and moves it through your pipeline. The Smart Files system stands out. Proposals, contracts, invoices, and questionnaires all use the same editor, which keeps the experience cohesive. Even small teams can deliver a polished client experience with very little effort.

Automation ties everything together. You can describe your process and let the AI build a draft workflow, start from a template, or assemble your own. The system feels approachable even if you have never created automations before.

Beyond that, HoneyBook folds in scheduling, tasks, time tracking, expenses, batch emails, contact management, and financial summaries. Integrations with Gmail, Google Calendar, Zoom, Canva, Slack, QuickBooks, Asana, and Zapier help reduce the number of tools you rely on daily.

Where it falls short

HoneyBook is designed for simplicity but the platform is yet not ideal for businesses with complex workflows. If your process branches in multiple directions or requires detailed internal routing, you may run into limits. Tools like Dubsado are more flexible for people who need deep customization.

The AI is helpful but not analytical. It supports communication and preparation but does not offer forecasting, strategic insight, or advanced reporting. If you want guidance on bigger business decisions, HoneyBook does not stretch that far.

The reporting tools focus on essentials. They show revenue, bookings, and lead performance but do not provide multilayered insights. And while HoneyBook supports small teams well, larger teams may find the structure too narrow for their needs.

What makes it different

What sets HoneyBook apart is the way it treats operations as something that should feel manageable rather than mechanical. Many business tools try to prove how powerful they are by giving you endless options. HoneyBook takes a different approach. It trims what is unnecessary and focuses on the parts of your day that actually shape how your business feels.

Instead of trying to trouble you with complexity, it focuses on consistency. The tool does not make your decisions for you, but it creates an environment where decisions are easier to make. Projects stay on your radar without demanding your attention. Clients move forward without needing constant reminders. The system gives shape to your work in a way that feels supportive rather than strict.

HoneyBook’s AI reflects the same restraint as the rest of the platform. It is not trying to run the business on your behalf. Instead, it pays attention to the moments where people often get stuck and adds just enough support to keep things moving. A quick draft to break the pause before replying. A meeting summary so you do not go hunting for notes. A gentle nudge when an inquiry is going quiet. These small interventions make the work feel lighter without changing how you work.

My take

HoneyBook understands the invisible weight of running a service business. It does not try to transform your workflow. Instead, it steadies it. It catches the tasks that usually slip through. It gives you a cleaner view of your week. It helps you stay consistent without requiring extra energy.

It is not for every business. Larger teams or operations with complex processes will likely outgrow it. But for freelancers, consultants, creatives, and small service teams, HoneyBook offers something rare. It brings clarity to the part of your business that often feels scattered.

And when the moving pieces finally sit in place, you realise how much easier the work becomes. HoneyBook helps you get there.

AI in Product

How Far Can Bubble’s AI Builder Really Take You?

I’ve been seeing more people talk about AI app builders lately, so I wanted to see how they hold up in day-to-day use. Bubble has been part of the no-code space for years, and people use it to build everything from simple tools to fully fledged web applications. It gives you a lot of freedom, but it also expects you to learn how the editor works before you can build anything meaningful.

The new AI builder is meant to soften that learning curve. You describe your idea, and Bubble puts together the first draft so you’re not starting from nothing. In theory, it should take you from an idea to something you can click through much faster.

To see how well that works in practice, I gave it a basic product idea that should be easy enough to translate into screens and workflows: a simple project management app with a clean dashboard and a few straightforward interactions.

What’s interesting

Bubble understood the brief and broke it down into features in a clear, structured way. The initial blueprint felt organised, and it captured the broad shape of the app I had in mind. It suggested pages, user flows and basic functionality that made sense for the concept.

The real insight came once Bubble started building the app inside the editor. You can actually see the AI working: adding elements, adjusting layouts and filling in placeholder content. It shows exactly what it’s doing, step by step.

That transparency makes it easy to understand how Bubble thinks about the task. But it also highlights the limitations. The AI follows the process, but it doesn’t always take initiative. It arranges components correctly, yet the output sometimes feels more like a draft of a draft rather than a near-ready interface.

This contrast becomes clearer once you compare it with Lovable later, which takes the same idea and produces something that feels more polished from the start.

Where it works well

Bubble’s AI builder is helpful when you want to avoid the blank-canvas problem. It gets you to the first version of your app quickly, especially if your idea has a straightforward structure. For example, a dashboard page, a tasks page and a project view all came together without much trouble.

It works best for:

• teams that plan to customise heavily

• people who already understand Bubble and want a head start

• early explorations where rough screens are good enough

Bubble also gives you full control once the AI step is done. You can adjust anything: spacing, components, workflows, data structure. For builders who like that level of control, Bubble’s flexibility is still its biggest strength.

Even so, some generations require cleanup. A few screens look tight and organised; others feel like they need alignment, spacing fixes or layout tweaks. The AI gets you moving, but you often need to refine what it produces.

Where it falls short

The gaps show up in consistency. Some elements sit too close together, others are spaced unevenly, and certain screens feel disconnected from the rest.

The AI also tends to assemble pieces rather than design a unified product. It places components correctly but doesn’t always think about flow, hierarchy or visual balance.

Another limitation is speed. The “building key features” phase can take longer than expected, and the output still feels early-stage.

Lovable, by contrast, approaches the same problem with more confidence. It commits to a design system, uses consistent spacing and produces layouts that look thought-through from the start.

Bubble’s AI helps, but it still expects you to be involved. Lovable helps by shaping the product more decisively.

What makes it different

Bubble’s AI layer sits on top of a mature editor. It doesn’t replace the tool; it works within it. That gives you transparency and full control, which is helpful if you like adjusting details or plan to extend the app deeply.

This also means Bubble’s AI behaves more like someone using the editor than someone designing the product. It doesn’t skip steps or make big decisions. It builds steadily, piece by piece, the way any user would.

Lovable, on the other hand, starts from intention. It decides on patterns, structure and layout before generating screens. It behaves more like a product designer than an editor operator.

My take

Bubble’s AI builder is useful, but it still feels early. It moves you past the blank screen and gives you a workable first version, but it doesn’t remove the need for refinement. If you like Bubble and prefer having control over every detail, the AI helps you get started faster without changing the way the tool works.

Lovable, though, feels like it understands the product itself. It creates layouts that already look like something you could put in front of someone, which makes the experience feel more complete.

If I wanted full control and expected to spend time editing, I’d use Bubble.
If I wanted to go from idea to a convincing prototype quickly, I’d use Lovable.

Bubble builds the draft.
Lovable builds the direction.

And depending on what you need, one may fit far better than the other.

AI in Design

Exploring 3D Worlds With Marble

A lot of early design work depends on understanding how a space feels. Static images can give direction, but they often miss how a room opens up, how lighting travels across it, or how elements relate when you look around. This is where 3D scenes become useful. They give you more context than a flat visual and make it easier to judge whether an idea holds together as an environment rather than a single frame.

Marble is part of this shift. You write a prompt or upload a reference image and it generates a scene you can rotate through. The output is closer to a quick spatial sketch than a finished environment, but it still shows how the tool reads depth, material and atmosphere. I could not test the 3D model upload because it is only available in the paid plan, but even with image inputs, Marble gives a clear sense of what it is trying to do. Its interface also helps. The Explore tab surfaces worlds created by other users, which makes it easier to understand what the tool handles well. The studio view lets you save, edit and organise scenes without switching tools. These small additions make the experience feel more guided.

What’s interesting

What stood out about Marble is how often it manages to produce scenes that feel surprisingly believable. In the greenhouse example, the lighting and plant textures carried through the full 360-degree view in a way that felt consistent with the photo I uploaded. You could rotate the scene and the mood stayed intact.

The Bohemian room example showed this even more clearly. The warm tones, the textiles and the overall layout translated well into a space that looked lived in. It was not perfectly stitched together, but the realism of the materials and the way the scene held its atmosphere made it easy to imagine someone standing inside it.

Marble’s presets also help. You can switch between realism, stylised and interior modes depending on what you want to explore. Each preset affects lighting and texture choices without breaking the overall feel of the scene. This flexibility, paired with the ability to browse community creations in the Explore tab, helps you understand what direction might work best for your own reference.

This level of realism sets Marble apart from tools that lean heavily into stylisation. Tools like Skybox (by Blockade Labs) often produce scenes that look artistic or abstract. Marble feels more grounded. It aims for spaces that resemble real interiors, which makes it more useful when you want something practical rather than playful.

The style controls help too. Switching between realism and stylised modes changes the mood without breaking the sense of space. This makes it easier to test different directions before choosing one.

Where it works well

Marble works well when you need to get a feel for a space without committing time to proper 3D modelling. It helps interior designers, art directors, illustrators and game concept teams gather ideas quickly. The scenes are not precise, but the realism is good enough to judge whether a direction feels appropriate.

The tool is also helpful for moodboards. Instead of stitching images together, you can generate a single environment that defines the mood more clearly. The panoramic view gives enough information about lighting, layout and textures to move conversations forward inside a team.

Marble suits people who enjoy exploring visual ideas and want a quick way to test environments. It asks very little upfront and gives you something concrete to react to.

Where it falls short

The limitations show up quickly. Without access to 3D model uploads, you cannot test how Marble performs with structured geometry. This is likely where the tool is strongest, but the free version does not let you explore that.

Even within 2D input, some scenes feel cohesive while others show stretched textures or repeated details when you rotate. The transitions between walls and objects are not always clean. If you need precision, Marble will not give you that. It gives you a direction, not a usable asset.

This is where tools like Meshy or Luma take the lead. They offer more control and produce output that is closer to real 3D work, but they also demand more from the user. Marble sits in a different spot. It is lighter, faster and more accessible, but less reliable when it comes to detail.

For teams that need consistency, Marble’s variability might be a challenge. Some scenes work immediately. Others need a few iterations before you find something usable.

What makes it different

Marble’s scenes look convincing at a distance, but they do not always hold up when you rotate through them. Some textures stretch, certain edges blur and a few surfaces repeat in ways that break the sense of geometry. These are expected limitations of tools that generate 360 degree environments from 2D inputs.

The free plan makes this more noticeable. Without 3D model upload, Marble has to infer the structure of the room from the prompt or image, which means the results depend heavily on the quality of the reference. Simple, well lit photos usually work better than complex compositions.

Other tools, like Meshy or Luma, produce more accurate structure because they rely on 3D reconstruction rather than panoramic generation. They offer cleaner depth and geometry, but they also require more steps and are not as quick for early exploration. Marble sits in a simpler category. It focuses on speed and accessibility, but that comes with tradeoffs in consistency.

My take

Marble is a helpful tool for early exploration, especially when you want to think in spaces without opening heavy software. It gives you something simple but useful: a draft you can rotate through, a sense of depth and mood, and enough information to decide whether an idea is worth developing.

It is not for technical 3D artists who need precision. It is not for people who want polished scenes. It is for designers who want to see how an idea translates into a space before they invest more time.

Marble will not replace 3D tools, but it fills a gap between imagining a room and building one. For early ideation, that is often the part that matters most.

In the Spotlight

Recommended watch: Marble — The First Controllable 3D World Model You Can Try Today

This walkthrough of Marble shows what controllable 3D generation looks like when it is actually usable. The video moves past abstract claims and shows Marble turning a single photo, a short prompt or a set of images into a navigable scene. Rooms, hallways, kitchens and even imagined interiors are reconstructed with enough realism to feel believable, and the editing tools make it possible to adjust the scene without breaking its consistency. The output still has limits and the edges show, but the overall experience feels closer to a real design tool than a tech demo. It is an early look at how world models might fit into everyday creative work once they mature.

“We now have the first multimodal frontier world model, fully controllable. A world model you can actually move around in.”

— Matthew Berman, host • ~0:00-0:08

This Week in AI

A quick roundup of stories shaping how AI and AI agents are evolving across industries:

DeepSeek goes head-to-head with GPT-5 - Chinese AI startup DeepSeek released new models (V3.2 and V3.2-Speciale) that claim performance on par with or better than top-tier models like GPT‑5 and Gemini 3 Pro on coding, reasoning and math benchmarks – all under an open-source licence.

Alibaba’s Qwen app becomes the world’s fastest-growing AI app - Qwen reportedly surged to over 149 million monthly active users (MAUs), making it the fastest-growing AI consumer app globally. Its rapid growth underscores growing user demand for accessible AI tools integrated into broader consumer ecosystems.

Google Workspace Studio brings AI agents to your everyday apps - Google’s new Workspace Studio surfaces agents inside Gmail, Docs, Drive and Chat – letting users build automations and workflows without code. It marks a shift: AI agents are moving from isolated assistants to integrated tools inside software many already use.

AI Fynds

A curated mix of AI tools that make work more efficient and creativity more accessible.

Fooocus → A simplified image-generation tool that lets you create high-quality visuals with clean prompts and minimal setup.

Meshy → A 3D creation platform that turns text or images into textured models, ready for design, gaming or product workflows.

AI Video Generator → A quick, browser-based tool for generating short AI videos, ideal for fast product clips or mood visuals.

AI Out of Office

Closing Notes

That’s it for this edition of AI Fyndings. From HoneyBook’s take on simple, supportive AI for freelancers and solopreneurs, to Bubble showing how an AI builder behaves inside a familiar product, to Marble turning simple inputs into explorable scenes, this week was about AI stepping closer to the starting point of how ideas become tangible.

Thanks for reading. See you next week with more stories, tools and perspectives shaping how intelligence continues to evolve across business, product and design.

With love,
Elena Gracia
AI Marketer, Fynd

Welcome to AI Fyndings!

Some shifts in AI arrive loudly. Others begin in places we barely notice. This week’s change sits inside the browser, the tool we use without thinking. It is starting to anticipate our intentions, highlight options and complete small tasks before we ask for them.

This edition is a little different. Once a month, we pause the weekly roundups and focus on one defining shift in AI, unpacking it with a little more depth and clarity.

In Perspective: Let’s look at why this shift matters and why the tension between Perplexity and Amazon signals the first real faultline in an internet where agents begin to make decisions before we do. Agentic browsing turns the browser from a passive doorway into an active guide, shaping what we see and how we move through the web.

By the way, since it’s Black Friday, a few of our AI products are offering some crazy deals this week. If you’re exploring new tools, these might be worth a look:

• Kaily (Conversational AI agents) → 25% off Growth and Business annual plans (use code BF25)

• Boltic (AI workflow automations) → Pay for 6 months and get 6 months free (use code BF50)

• GlamAR (AR try-ons) → 15% off on all plans (chat with sales to redeem)

• Pixelbin (AI image editor + text-to-video) → 30% off Lite and Pro plans (use code BF30)

• Upscale.media → 30% off annual subscriptions (use code BF30)

We have spent so much time discussing what AI can do for apps and tools that we’ve almost ignored the place where everything actually starts. Not the model, not the app, not the workflow. The browser.

For years, the browser has been the most simple part of the stack. You open it, search for something, click through a few links, compare a handful of options, and eventually find what you were looking for. It doesn’t actively guide the journey or filter the choices. It just keeps the pages in front of you.

That mental model is breaking. Slowly giving way to something new.

You can see signs of this in the day-to-day interactions most of us overlook. You search for a product and instead of juggling ten tabs, the browser brings back a few strong options, sometimes with a clear “best pick” based on price, reviews and delivery time. You look for a place to stay and it quietly pulls from Airbnb, Booking.com and Expedia, cleaning up the noise into something you can skim in seconds. You try to track a package or renew a subscription and the right status or form fields appear before the page has fully loaded.

At first this looks like a natural evolution of “smart features”. We already have auto-fill, ad blockers, extensions that compare prices and search results that summarise half the web for us. None of that feels like a turning point. It’s still you, clicking through, making choices, doing the work.

The important shift is more subtle.

When people talk about agentic browsers, they’re not talking about a slightly smarter Chrome. They are talking about a browser that can listen to an intent like “Find the best monitor under ₹20,000 that ships this week and buy it” and then handle the messy steps in between, from discovery to purchase.

With traditional browsers, your flow looks like this:

You think of a goal → you search → you scan → you click → you compare → you decide → you act.

With the agentic version, it looks more like this:

You state the goal → the browser interprets → it searches, scans, clicks, compares, decides → and then it presents you with the final product choice or a ready-to-checkout page.

The effort hasn’t disappeared. It has just moved. The decision-making process shifts from your hands to the logic of an agent.

Tools like Perplexity’s Comet, Atlas, Fellou and others are explicitly built around this idea of turning the browser into an “action engine” instead of a passive window.

So what is truly new here?
Not that tasks are automated. That has existed for years in the form of scripts, RPA and individual apps.

The difference is that the agent sits at the browser layer, understands intent in natural language and can work across almost any site you use, from ecommerce to travel to admin tools, without each of those sites designing a special integration first.

This is the backdrop that makes the Amazon and Perplexity conflict so telling. What looks on the surface like a policy violation is actually the first real friction between two very different ideas of how the web should work.

Perplexity’s Comet browser takes the agentic idea literally. You tell it what you want, and it doesn’t wait for you to click through Amazon like a regular shopper. It navigates for you. Comet logs into your account, looks at products, compares sellers, weighs delivery times and can even move toward checkout, all while appearing to Amazon like any other Chrome user browsing the site. To Perplexity, this is not a loophole. That’s the entire point. A browser agent should behave like an extension of the user, not a passive observer.

Amazon has an entirely different perspective on this.

Their retail media business depends on the idea that people browse. Not briefly, not efficiently, but deliberately and repeatedly. Sponsored listings, recommendation rails, comparison grids, cross-sells, delivery bundling, Prime nudges are all designed for a human moving through the interface step-by-step. When an agent slices through that flow and jumps straight to the “best option,” Amazon doesn’t just lose a click. It loses the surfaces that make the business work.

So when Perplexity’s browser began completing transactions on Amazon without flowing through the usual steps, the conflict surfaced quickly.

Amazon argued that the agent was bypassing essential layers of the experience and that it cannot guarantee trust or safety if an external system manipulates its flows. Perplexity argued that the user should decide how they prefer the journey to unfold. What appeared on the surface as a policy disagreement was, in reality, a clash between two visions of the web. One sees the agent as the primary guide. The other sees the platform as the central gateway.

Behind the statements, the disagreement is simple: Who should be in charge of the journey? The platform or the agent?

For Amazon, the journey is the product.

For Perplexity, the journey is unnecessary friction.

And this is where the conversation becomes bigger than two companies. If agentic browsers take off, they chip away at the assumption that users will dutifully follow each platform’s designed path, including the filters, the sort options, the comparison modules, the ads, the nudges, the endless “You might also like.” Those steps were built for a world where the user is present at every click.

An agent isn’t present in the same way. It sees those layers as backend details. Its job is not to browse but to deliver an outcome.

For big platforms that rely on people browsing, for example Amazon, DoorDash or Ebay, agentic browsers feel threatening. These platforms are built on the idea that users will look around, compare options and spend time around the experience. That is where ads, recommendations and upsells/cross-sells live. When an agent skips that entire flow and goes straight to the final step, the platform loses the chance to influence the decision. The Verge calls this the “DoorDash problem,” where an agent treats every platform as just a delivery system in the background, not a place with its own design, incentives or brand.

For smaller brands, the impact is a mix of good and challenging. On one hand, an agent that truly works for the user might consider sellers that it would normally take several scrolls to discover. Which means smaller players could actually get more visibility if they offer better pricing, faster delivery, or when their product matches the user’s needs more precisely. This could be the exact shade of olive green a user searched for instead of the generic khaki at the top, the wide-strap sandal the user actually wants rather than the narrow version at the top of the page, or the soft cream fabric chair instead of the bright white one favored by the algorithm. But on the other hand, the competition doesn’t disappear. It simply shifts. Instead of trying to appear on the first page of a platform’s search results, brands may end up trying to appear in an agent’s shortlist of “trusted” or “reliable” options, a list the user might never see or fully understand.

That shift is already visible in how different competitors respond. Alibaba’s AI Mode embraces agents rather than resisting them by integrating agentic flows directly into the marketplace, where early data shows substantial gains in orders and supplier activity. OpenAI’s Atlas and Fellou approach it from the consumer side, building browsers that can act across logged-in sites. Microsoft’s FARA-7B sits even deeper, operating the entire computer interface instead of just the web layer.

Taken together, these developments point to the same idea: the browser is turning into a worker, not a window.

There are real advantages to agentic browsing. For everyday tasks, such as comparing prices, planning a trip, renewing a subscription or checking a delivery, an agent can remove a lot of the small steps that slow us down. It makes the internet feel easier to use, especially for people who find complex interfaces confusing or don’t speak the platform’s primary language. And for businesses, especially in operations, sourcing and internal workflows, these agents can handle the manual work that usually takes hours. They can review documents, compare suppliers, summarise contracts, coordinate schedules and pull data from scattered tools.

But more autonomy also means more places where things can go wrong. A browser that can act for the user can also be misled. Some early studies have shown how malicious websites can trigger unintended actions by feeding misleading instructions to the agent. A browser that autofills forms can accidentally share sensitive details. An agent that completes a purchase could be nudged toward an untrustworthy seller because the page structure subtly manipulates its inputs. And even when nothing malicious is happening, there’s still a question of judgement. When should an agent step in and act for you, and when should it stop and ask first?

This is not theoretical. Security teams at major companies are starting to warn that agentic browsers expand the attack surface. Traditional browsing limits the damage because a human still reviews each step. With agentic behaviour, the browser may act first and the user only notices later. The same autonomy that streamlines the experience also creates room for mistakes, misinterpretation and manipulation.

As these tools grow, the industry will need safeguards, clearer permissions and transparent logs that show why the agent made a particular choice.

Microsoft’s FARA-7B model tries to solve this by adding “critical points” – moments where the system must pause and get your approval before doing something important. Agentic browsers will need their own version of this, or they risk acting too quickly on things the user would have handled more cautiously.

Another concern is transparency. When an agent chooses one seller over another, we need to know why. Was it the price? Delivery? Reliability? Partnership incentives? Something in the training data? Without clarity, we risk replacing platform-driven influence with agent-driven influence without understanding either.

Which brings us back to Amazon and Perplexity. Their conflict isn’t about Chrome spoofing or terms of service. It’s more about who gets closest to the user’s intent – the platform or the agent.

I see this moment as a turning point. Not because of who is right, but because of what the conflict represents. We are moving into a phase where the most influential part of the online experience may no longer be the platform or the content, but the agent interpreting both.

If agentic browsing continues to grow, companies will need to rethink how they reach users and how they build trust. Users will need to understand how their agents decide and what those decisions are based on. And somewhere in the middle, new rules will need to emerge to protect fairness, choice and transparency.

For now, we are only at the beginning. But the Perplexity and Amazon clash shows how quickly this shift is becoming real. When the browser starts thinking for the user, the stakes rise for everyone involved. The next few years will show whether agentic browsing becomes a helpful companion, an invisible filter or the new gatekeeper of the web.

In the Spotlight

Recommended watch: Perplexity vs Amazon: The AI Agent War Begins

I found this episode really interesting as it makes the Amazon and Perplexity clash easy to understand. It highlights the important details, from Amazon accusing Comet of posing as Chrome and bypassing things like delivery bundling and curated shopping signals, to Perplexity’s CEO pushing back with his now public “Bullying is not innovation” letter. It makes the tension clearer and shows how agentic browsing is beginning to challenge the logic behind large platforms.

“AI agents have an adverse effect on advertising. It all leads to a return on ad spend that’s very unusual.”

– Amazon CEO, as quoted in the episode • ~1:34

Closing Notes

That’s it for this edition of AI Fyndings. From agentic browsers starting to shape our online decisions, to Amazon and Perplexity revealing the first real tension between platforms and AI agents, to new tools experimenting with autonomy across search, commerce and everyday workflows, this week was about intelligence quietly moving into the spaces we once assumed were neutral.

Thank you for reading. See you next week with more stories, tools and ideas shaping how AI continues to redefine how we explore, choose and interact with the digital world.

With love,
Elena Gracia
AI Marketer, Fynd

With AI, every decision is a trade-off: speed or quality, scale or control, creativity or consistency. AI Fyndings discusses what those choices mean for business, product, and design.

In Business: Lightfield turns the CRM into a living memory that builds itself.

In Product: Maksai experiments with bringing automation into WhatsApp’s familiar chaos.

In Design: Affinity by Canva unifies design into one space, challenging Adobe’s rulebook.

AI in Business

Do We Still Need To Update CRMs Ourselves?

Every team I’ve worked with hits the same wall with their CRM.

You start with optimism, thinking this time the data will stay clean and up-to-date. A few months later, the pipeline is half-empty, nobody logs calls, and you’re back to running your business on Slack threads and memory.

Then I found Lightfield, an AI-native CRM that claims to change that dynamic. It promises to build itself by capturing everything your team does automatically. I have not used it yet, but I spent time reading through early reviews, product notes, and research write-ups to understand the idea behind it. What I found feels like the beginning of something new in how businesses manage relationships.

What’s interesting

Lightfield’s premise is simple: instead of asking people to enter data, it listens to what is already happening inside your company. It connects to your email, calendar, support tools, and even product analytics. It then stitches those signals together into what the founders call a “living memory” of your customers.

In practice, that means you can ask questions like “Which accounts are most engaged this quarter?” or “Who has not been contacted in the last month?” and the system can answer based on real data. It is the type of shift that feels small in concept but huge in impact.

Traditional CRMs were built for reporting. Lightfield is built for awareness. It wants to help you understand what is happening right now, not what happened last month.

Where it works well

Lightfield appears to work best for smaller or early-stage teams where customer activity is high but discipline around documentation is low. Think founder-led sales teams, small go-to-market functions, or startups where people wear multiple hats and nobody has time to update a CRM.

Users describe it as freeing. It captures all those scattered interactions automatically. That alone could save hours each week and, more importantly, preserve context that often gets lost between calls, handoffs, or channels.

The interface lets you query data in plain English, making insights accessible without reports or dashboards. Teams that have outgrown spreadsheets but find Salesforce too heavy might find this refreshing.

For fast-moving teams, Lightfield looks like the first CRM that actually keeps up.

Where it falls short

Of course, this kind of automation has limits. Lightfield is still new. Some users describe it as promising but unfinished, and the product is clearly evolving fast. The integrations are improving, but the experience depends on how much of your workflow is already digitized.

It performs best when it can tap into a steady stream of structured and unstructured data. If your team’s customer communication happens across disconnected tools, Lightfield may not see the full picture.

The company has also been teasing an Open Platform that will let developers and teams build their own integrations and workflows directly inside Lightfield. Once it launches, the platform could open up much more flexibility and make it easier to customize Lightfield to different business types.

And while the idea of “AI that acts for you” sounds exciting, it also introduces a trust factor. Many teams still prefer to review what goes out to customers, and automation requires a level of comfort with letting the system make decisions on your behalf.

The competition does not make it easier. Hubspot continues to evolve, layering in its own AI features while staying accessible to small and mid-sized businesses. Salesforce remains the enterprise default, with deep integrations and customization that Lightfield cannot yet match. At the same time, newer CRMs such as Attio, Folk, and Clay are changing expectations. They are lighter, more collaborative, and designed to fit naturally into the tools teams already use. These products shift the focus from rigid data entry to shared context, showing that the future of CRMs will be about adaptability rather than scale.

Lightfield sits in an emerging middle ground. It is lighter, faster, and more intelligent, but also younger and less proven.

What makes it different

Lightfield stands out for how it treats information. Most CRMs focus on structured fields such as deal value, close data, and stage. Lightfield, on the other hand, pays attention to the conversation itself. It reads the text, tone, and content of meetings and messages. Then turns that into context.

It was built from the ground up for AI. This is important because it means the intelligence is not a bolt-on assistant sitting beside the system. It is the system. Every workflow, recommendation, and reminder comes from how the AI interprets your activity.

This design makes Lightfield feel less like a database and more like a real-time record of your customer relationships. It does not just store data but tries to connect dots.

My take

I find Lightfield genuinely interesting. The ideas of a CRM that works in the background while you can focus on customers feels overdue. It recognizes that most people do not hate CRMs but the manual work that tags along.

If it can actually deliver a reliable, trustworthy experience, it could redefine what “sales software” even means. But it still has a long way to go. The Open Platform feature needs to launch, integrations need to deepen, and the automation will have to earn users’ trust.

If I were running a small sales or success team today, I would pilot it. I’d start with one vertical, connect my email and calendar, and see if it can actually replace my manual notes. The test would not just be about saving time but about whether the insights it produces are better than what your team already knows intuitively.

Salesforce remains the old empire. Hubspot is still the safe ground. Lightfield is the young, ambitious challenger that believes CRMs can think. Whether that belief turns into a product that scales will depend on how well it executes in the next year.

And honestly, in a category that has not seen real change in decades, that is exactly what makes it worth trying.

AI in Product

What Happens When WhatsApp Gets an Assistant?

I tried something different this week. Instead of another AI workspace or productivity app, I tested an assistant that works entirely inside WhatsApp. It is called Maksai, and it claims to help manage messages, calls, and reminders directly from chat.

The idea sounds perfect. A lightweight assistant that fits into a tool I already use every day. No new logins, no new interface, no dashboards. I was curious to see how far an assistant inside WhatsApp could go in actually helping me work more smoothly.

So I signed up, linked my number, and started chatting.

What’s interesting

At first glance, Maksai feels promising. It shows up as a contact on WhatsApp. You can use it on your phone or through your browser, and the onboarding is quick and simple.

Maksai introduces itself as your personal assistant that can set reminders, schedule messages, mute chats, send auto-replies, and alert you about important messages. It also says you can connect Google Calendar so that it can help schedule meetings and block time, though I could not confirm how deeply that integration works beyond what the company claims.

On paper, the idea makes sense. Most of our work conversations already happen on WhatsApp, so having an assistant there feels natural. It positions itself as a quiet helper that can take care of small actions while you focus on bigger ones.

That is where the potential ends and the problems begin.

Where it works well

To be honest, there is not much here that works. Maksai’s onboarding is smooth, but that’s about it.

It responds quickly to greetings, acknowledges commands, and confirms tasks, but nothing happens after that. No reminders. No alerts. No scheduled messages. Every feature exists only in words.

It gives the illusion of progress through common indicators and polite confirmations, but nothing actually happens after that. The interface itself is approachable, and you don’t need to learn prompts or switch tools, but accessibility without real outcomes doesn’t add much value.

Where it falls short

Everywhere else.

When I asked it to alert me whenever a contact texted, Maksai immediately replied, “Got it.” But the alert never came. I tried setting a reminder next; it confirmed the task, then stayed silent. It can acknowledge almost anything you ask but never follows through.

It also does not recognize contacts or groups unless new messages come in. It does not access or understand existing chats. The response time is average, but since no actions follow, speed hardly matters.

Maksai appears as a separate WhatsApp number rather than an integrated feature. For a platform as personal as WhatsApp, that feels odd and a little concerning. When an unknown number claims to be your assistant, trust becomes part of the product experience.

Then there is the question of privacy. Allowing a third-party assistant to read and process messages is not an easy decision. Maksai says all messages are deleted once processed, but there is no way to verify what that means in practice. Even if the system is secure, the idea of giving an external AI access to private conversations is concerning.

The comparison becomes clearer when you look at Meta AI, which is now being integrated directly into WhatsApp. Meta AI can answer questions, summarize information, and provide assistance within the app itself while operating inside WhatsApp’s own infrastructure. Maksai, on the other hand, sits outside that ecosystem. It promises similar help but without the same trust, security, or reliability that come with being a built-in feature.

What makes it different

The idea of building an assistant inside WhatsApp is clever. It is where people already spend most of their time, and it lowers the barrier to trying something new.

But that familiarity also works against it. WhatsApp users expect things to feel seamless and safe. When an assistant appears as an unfamiliar number, fails to deliver on its promises, and cannot clearly explain what happens to their data, it becomes difficult to trust. Maksai’s concept is interesting, but its design and privacy model make it feel unfinished.

At this stage, it feels more like a prototype than a product.

My take

After testing Maksai, I can say it is not ready yet. The idea makes sense, but the execution is missing. It is a conversation shell that acknowledges instructions without taking a single action.

If the vision is to create an assistant that truly lives in chat, Maksai still has everything left to build: functionality, reliability, and transparency. And while the team claims to delete all processed data, privacy inside a platform like WhatsApp is not something that can be taken lightly, especially when it involves sharing information with an external number.

With Meta AI now offering similar capabilities natively within WhatsApp, the contrast is clear. One works within a trusted ecosystem and delivers; the other sits outside it, still learning how to keep its promises.

Right now, Maksai serves as a reminder that good ideas need solid execution to earn trust. It sits at the intersection of messaging and automation but has yet to take its first real step forward.

If I had to describe it simply: Maksai is what happens when products promise action but stop at conversation. The space it aims to occupy is exciting, but this version does not deliver enough to belong there yet.

AI in Design

Affinity by Canva vs Adobe: Redefining the Designer’s Workflow

When Tejas, one of our designers, opened the new Affinity by Canva, his first impression was that it looked fast, clean, and ambitious. The interface is stripped of clutter and looks ready for work. But could it really hold up against Adobe, a suite that has shaped professional design for decades?

Affinity’s update merges what used to be separate apps – Designer, Photo, and Publisher – into one unified software. You can now switch between pixel editing, vector design, and layout work without ever leaving the canvas. It is free to use, with its AI features available through Canva Pro.

What’s interesting

The most noticeable change in Affinity is how fluid it feels. Instead of switching between multiple programs, you work in one space where every creative task connects. You can edit a photo, tweak typography, and export a layout without opening new files.

Tejas noticed how color correction tools feel inspired by Adobe Lightroom. You can fine-tune tone curves, contrast, and highlights without bouncing between panels. There is even an option to connect Google Calendar to schedule creative tasks, which shows how Canva wants to merge design and workflow in one place.

The other thing that stands out is speed. Affinity opens quickly and handles large files smoothly. Compared to Adobe’s heavier software, which often takes time to load and sync, Affinity feels light and responsive.

Where it works well

The single-software design is Affinity’s best feature. In Adobe’s ecosystem, you need Photoshop for photo work, Illustrator for vector art, and InDesign for layouts. In Affinity, those boundaries disappear, creating a sense of continuity that helps you stay in the creative flow.

Tejas liked how well the Slice, Vector, and Tone Mapping tools performed. They were quick, accurate, and easy to use.

Affinity also brings several creative tools powered by Canva. You can remove backgrounds, fill empty spaces, upscale images, enhance colors, and generate visuals directly within the same window. These tools are fast and useful for brainstorming or building early concepts, though most of them still require Canva Pro to unlock.

The price difference alone makes Affinity appealing. It offers professional capabilities for free, while Adobe’s suite requires a monthly subscription. For independent designers or small teams, that cost gap is significant.

Where it falls short

The gaps begin to show when you expect Adobe-level refinement.

​​Generative Fill, powered by Firefly, remains one of Adobe’s strongest advantages. It rebuilds missing pixels, extends backgrounds, and adjusts lighting with near-perfect accuracy. Affinity aims to offer similar capabilities, but its free version stops short of advanced automation. The AI-assisted features, available only through Canva Pro, are still developing and can feel less consistent than Adobe’s. This makes Affinity’s free tools strong for precision work, but not yet a match for Adobe’s automated refinements.

File compatibility is another concern. Opening PSD or AI files in Affinity can cause alignment or effect issues. For teams who have built years of archives in Adobe’s formats, this adds unnecessary friction.

Some tools that work instantly in Adobe take a few extra steps in Affinity. Basic actions like refining edges, color matching, or aligning objects are not as fluid. The difference is small but noticeable when you work at scale.

And then there is adoption. Adobe has been around for over three decades. Designers have grown up learning its shortcuts, file systems, and behaviors. Affinity may be easier to learn, but it is hard to unlearn a language the industry already speaks fluently.

What makes it different

Affinity succeeds in offering a fresh, simpler way to work. It removes layers of complexity and focuses on speed, making it ideal for designers who want less overhead. You can move from sketch to export without leaving one environment.

Canva’s influence shows up in how these tools feel approachable, even for beginners. The creative features are well-integrated and do not sit behind plug-ins or menus. They may not be as refined as Adobe’s, but they are designed to be usable by anyone.

Affinity prioritizes accessibility and simplicity, while Adobe still leads when it comes to depth and precision. They are built for different audiences: Adobe for professionals who need total control, and Affinity for those who value flexibility and ease.

My take

After a week of testing, Tejas called Affinity “a fresh take that feels light, fast, and almost there.” I agree.

Affinity by Canva is the most promising alternative to Adobe in years. It is quick, thoughtfully designed, and genuinely fun to use. But it still feels like a product in progress. The creative tools are helpful but uneven, and Adobe’s polish, precision, and consistency remain unmatched.

Even so, Canva’s move into professional design software is a big moment. It challenges Adobe’s monopoly and introduces an inclusive approach to creative work, one that values access over complexity.

Affinity is not a replacement yet, but it points toward a new direction for design software: simpler, smarter, and open to everyone.

In the Spotlight

Recommended watch: Waveform Podcast - The Future of Free Software

In this week’s episode, Marquez Brownlee and his team dive into one of the most surprising moves in creative software: Canva making Affinity completely free. The conversation unpacks how Canva’s acquisition is shaking up Adobe’s dominance, what this means for designers who’ve spent years inside Creative Cloud, and whether “free forever” really ever means forever.

Canva Premium is about $13 a month or $120 a year. Adobe Creative Cloud Pro can run over $70 a month. Affinity gives you the core tools for free – even half the price of one Adobe app.
– Andrew, Waveform Podcast • ~15:50

This Week in AI

A quick roundup of stories shaping how AI and AI agents are evolving across industries:

Gemini 3 by Google now introduces “Deep Think” mode and full multimodal reasoning across text, image, video and code, signaling a shift from tools that help to tools that really understand.

Manus AI launches the Browser Operator extension, letting an AI agent live inside your browser and automate tasks across logged-in apps. It signals a shift from “AI answers questions” to “AI does things.”

Ai2 releases Olmo 3, a new generation of open-source models that focus on reasoning and logic instead of pure scale. The release marks an important step toward more transparent and explainable AI systems.

AI Out of Office

AI Fynds

Napkin AI → A tool that transforms your text into diagrams, infographics, and visuals instantly.

Flow → A motion-design app that turns text and shapes into smooth animations, bridging quick sketches and polished motion graphics.

YouTube Thumbnail Generator → An easy and quick way to create sharp, high-contrast thumbnails that stand out without opening heavy design software.

Closing Notes

That’s it for this edition of AI Fyndings. From Lightfield rebuilding the CRM as a self-learning system, to Maksai testing how automation fits into everyday chat, to Affinity by Canva proving that professional design can be both fast and free, this week was about technology learning to simplify the way we work.

Thank you for reading. See you next week with more stories, tools, and ideas shaping how intelligence keeps finding its place in how we build, communicate, and create.

With love,
Elena Gracia
AI Marketer, Fynd

With AI, every decision is a trade-off: speed or quality, scale or control, creativity or consistency. AI Fyndings discusses what those choices mean for business, product, and design.

In Business, Hiro brings logic and structure to financial planning, turning complex decisions into something understandable.

In Product, Kimi K2 reveals what happens when an AI slows down and reasons out loud, proving that transparency is its own kind of intelligence.

In Design, Kite.video makes communication the new creative skill, helping products speak clearly when time and resources run short.

AI in Business

Hiro: When Your Personal CFO is an AI

Financial planning isn’t really about money. It’s about clarity, and that’s the part most of us never quite master.

Most people plan their finances in fragments: a budget here, an investment there, a vague idea of what “someday” should look like. Hiro wants to replace that patchwork with a single, evolving system that ties it all together.

It calls itself an AI financial planner, but that description undersells it. Hiro behaves like a personal CFO. It tracks your income, expenses, savings, debts, and goals, then builds a live model that adjusts as your circumstances change.

When I tried it, I wanted to see if that promise held up. So I gave it a plan that was intentionally flawed which included a modest income, a big house, a small down payment, and a short loan term. Within seconds, Hiro responded with a projection that said, very plainly, I’d run out of liquid assets by 2026.

It didn’t soften the message or try to be motivational. It simply showed me the numbers.

Then I asked it to fix the plan. Hiro adjusted the parameters, extended the loan, and resized the home value. The new plan looked realistic and achievable. What I appreciated was how it explained why the new structure worked. It wasn’t just solving for affordability; it was helping me see cause and effect.

That’s when I realised what makes it interesting. It’s not about making money management easier. It’s about making it make sense.

What’s interesting

What caught my attention about Hiro is how it structures your thinking. It doesn’t hide behind fancy dashboards or vague advice. It shows its reasoning. Every assumption is laid out – how much of your income should go toward housing, when liquidity becomes a concern, and how small changes can ripple through your long-term plan.

The experience feels conversational in the best way. I could ask, “What if I bought a home next year?” or “What if I took six months off work?” and Hiro would start building a scenario right in front of me.

What impressed me most was how much Hiro does beyond planning. Its Net Worth view gives a living snapshot of assets and debt, the Cash Flow section reveals spending patterns I hadn’t noticed, and Goals connects short-term choices to long-term outcomes. The Optimize feature even scans connected accounts to suggest where idle cash could earn more or how to get the best out of existing credit cards.

Where it works well

I initially thought Hiro was built only for big decisions, but it’s just as capable in the everyday. Its Cash Flow section turns daily spending into something you can actually understand. Each transaction folds into the bigger picture, so budgeting doesn’t feel like micromanagement.

It is also best in scenarios where you’re thinking about decisions with long-term consequences. It doesn’t feel like it is designed only for quick tips or tracking bills but more for the bigger questions, such as buying a home, switching jobs, planning a sabbatical, or trying to understand what “financial independence” really means in numbers.

Where it falls short

Hiro gets a lot right, but it still feels like a tool in progress. While it connects to your accounts and pulls real-time data, it doesn’t yet automate deep analysis. You can view spending, track patterns, and project outcomes, but you still need to interpret many insights yourself. It doesn’t take that next step of translating data into direct, actionable advice.

Forecasting is helpful but sometimes idealistic. Markets shift, inflation changes, and personal spending habits evolve, yet Hiro assumes a level of consistency that real life rarely provides. It gives you a solid foundation, but not a complete safety net.

There are also moments where the analysis feels too surface-level for advanced users. It highlights what is happening, like your savings rate dropping or your debt ratio rising, but it doesn’t always explore why or what should change first.

And despite its thoughtful design, Hiro still needs your participation. It works best when you connect accounts, verify transactions, and update information regularly. If you leave it unattended, its insights lose accuracy quickly.

Hiro is a smart planner, but not yet an adaptive one. It helps you see your financial life clearly, but you still have to decide how to act on it.

What makes it different

What sets Hiro apart is that it doesn’t just show you numbers but helps you understand them.

Most financial tools describe the past. Hiro models the future. It combines tracking, forecasting, and reasoning into a single system that behaves more like a strategy model than a budget app. You can simulate major life decisions, test assumptions, and watch how your net worth or cash flow changes instantly.

When I compared it to other tools, the difference was clear. Cleo focuses on personality and tone. INDmoney excels at asset aggregation. Monarch Money wins on design polish. Hiro, though, focuses on comprehension. It assumes you’re capable of thinking deeply about money and gives you the space to do it.

The best part is that it feels designed for people who already think in larger systems – founders, leaders, professionals who build strategies for a living. Hiro simply applies that same strategic discipline to personal finance. It’s not about being told what to do. It’s about learning how to reason through trade-offs.

My take

Hiro doesn’t try to impress. It tries to make you pay attention.

It’s one of the few AI tools that helps you see your financial life as a system instead of a collection of scattered choices. It doesn’t promise to make you rich or automate every decision. It simply gives you a clearer view of how your money moves and what it means for the future.

There’s still work to do. Hiro could go deeper in its analysis and offer sharper, more actionable insights. But even now, it does something that most tools don’t. It makes you stop, think, and understand where you actually stand.

AI in Product

Kimi K2 vs GPT-5: Slower, Messier, and Maybe Smarter

Kimi K2 is Moonshot AI’s latest large language model, built to handle complex reasoning, long-form analysis, and code generation. It’s designed to go beyond quick answers and show how it arrives at them, breaking problems into steps, tracing its logic, and explaining its reasoning along the way.

I wanted to see how it performs in practice, so I tested it against GPT-5 on three different tasks:

1. Building a simple game

2. Analysing AI startups, and

3. Designing an autonomous assistant.

The goal wasn’t just to compare accuracy but to understand how both models reason.

Kimi K2 was slower and more deliberate, often taking longer to form a complete answer. GPT-5, by contrast, was faster and cleaner, with less friction in its process. But where GPT-5 hides its reasoning behind polish, Kimi K2 exposes the scaffolding and sometimes that transparency is as revealing as the result.

1. Code generation

The first test was simple: “Create a Tic-Tac-Toe game for an HTML website.”

Kimi K2 built playable code that worked without modification. It explained each part clearly and described how a user could interact with it. The problem was speed. It took almost twice as long as GPT-5 to produce the same output.

GPT-5’s code felt ready to use – compact, styled, and tested. It even added keyboard control and a computer opponent. Kimi’s version was competent but mechanical. It worked, but it didn’t feel confident.

Kimi can build. GPT-5 can build well.

2. Research and reasoning

Next, I asked both models to analyse five emerging AI startups in Europe and rank them by differentiation and growth potential.

Kimi K2 approached the task like an analyst in training. It produced a ranking, scoring table, and commentary on each company’s market position and funding history. It even cited sources such as EU Startups and Financial Times. The detail was impressive, though not always consistent. A few references were plausible rather than precise.

GPT-5, in contrast, wrote less but said more. It chose fewer companies, presented its logic in a crisp structure, and avoided redundancy. Kimi liked to show its work; GPT-5 focused on the conclusion.

If you’re researching a topic from scratch, Kimi’s depth helps. If you need something clear enough to present, GPT-5 would be a better choice.

3. Systems design

For the third task, I asked both models to design an AI assistant capable of running a small online business.

Kimi K2 treated it like a full architecture exercise. It mapped data flow, reasoning loops, and safety mechanisms, even defining a failure mode it called “inventory hallucination.” It described how to detect and recover from it using a triple-check rule.

GPT-5 outlined the same system in less detail but more structure. It captured the key principles, delegation, feedback, and human override, without getting lost in its own logic. Kimi’s design was longer, more technical, and at times overengineered. But it was also imaginative, the kind of thinking that hints at potential beyond the surface.

What’s interesting

Kimi K2 is built for reasoning, not speed. It doesn’t aim to sound human; it aims to think clearly. It walks through logic, checks itself, and builds arguments one layer at a time. You can follow its trail, which makes it easier to understand how it got to an answer. GPT-5 usually hides that layer of process beneath fluency.

This makes Kimi a good model for exploration, when you want to see how ideas evolve rather than just get the final conclusion. It’s particularly useful in complex or creative workflows where reasoning transparency matters more than speed.

The recent posts by Alex Finn and Omar Saravia highlight this strength. Finn showcased Kimi’s ability to handle reasoning over large contexts, while Saravia pointed out its agent-like structure and self-verification loops. Both reflect what I saw: a model that prioritises logic, even when it occasionally loses focus.

Where it falls short

Kimi K2 often struggles to decide what deserves attention. It sometimes spends paragraphs explaining simple ideas and then rushes past the ones that actually need depth. It’s not that it gets things wrong; it just doesn’t always know what’s important.

This inconsistency makes it unpredictable. In one session, it might reason deeply about a minor point; in another, it skips essential context. The result feels uneven. They are thoughtful one moment, confusing the next.

The model is also noticeably slower. On coding and design tasks, GPT-5 finishes in half the time. Kimi takes longer to process and occasionally repeats itself while “thinking out loud.”

And while its reasoning is clear, its accuracy isn’t always reliable. It builds structure beautifully but still hallucinates details when summarising facts or referencing data. It’s the kind of intelligence that works hard but still needs editing.

Kimi’s challenge isn’t that it lacks ability; it’s that it hasn’t yet learned restraint.

What makes it different

Kimi K2’s difference lies in its transparency. It shows its reasoning in a way most models don’t. You can see its logic unfold: the checks, the doubts, the course corrections.

That visibility gives it personality. GPT-5 feels certain; Kimi feels earnest. One delivers, the other deliberates.

My take

Kimi K2 is not the most capable model I’ve used, but it’s one of the most revealing.

It’s slower, less consistent, and at times confusing, but it has a kind of effort you can sense. It wants to reason properly, even when it doesn’t. You can see the thinking, the hesitation, the correction.

For now, GPT-5 still leads on speed, accuracy, and clarity. But Kimi K2 is quietly exploring a different path, one that values transparency over perfection. It’s not a finished product, but it’s an honest one.

And in an industry obsessed with polish, that honesty feels like progress.

AI in Design

Kite.video: A Shortcut for Product Storytelling

Design often sits at the end of the process. You build, test, refine and then scramble to make it look good enough to share. Kite.video exists for that last part. It’s a desktop tool that helps you turn product ideas into short, launch-ready videos without needing motion design skills or editing software.

You open the app, pick a layout, and it already feels like a storyboard that’s been thought through. Add visuals, write a few lines, pick a soundtrack or voiceover, and you’re done. The app handles transitions, timing, and polish, giving your product a clear, structured narrative.

When I tried it, I didn’t overthink it. I chose a template, dropped in visuals, added text, and played around with the music options. It took me around 7 minutes to make the video embedded below, and it turned out way better than I expected.

That’s what Kite gets right. It makes your product look like it is meant to be seen.

What’s interesting

Kite has a clear design style with clean layouts, simple motion, modern fonts, and soft gradients. It is not trying to be flashy. It is trying to look professional with minimum effort.

The templates follow a steady rhythm with an intro, feature highlight, and closing frame, and the pacing feels right. The music options are small in number but well chosen. A calm acoustic track makes a demo feel like a tutorial, while an electronic one gives it energy. Those subtle choices make a bigger difference than you expect.

What I liked most is how it adds rhythm to static visuals. Even with a few screenshots, the output looks intentional.

Where it works well

Kite is great for short and direct videos such as product demos, feature launches, or quick internal updates. It works best when you want to explain something clearly rather than create something new.

Compared with Runway or Pika, Kite is simpler and faster for real product footage. It is more visual than Descript and it does not try to be a storytelling tool like Synthesia. It is a practical choice for small teams or founders who need design output but do not have a motion designer.

Where it falls short

Kite’s strength, its structure, also limits it. The templates are fixed, so every video ends up looking similar. After a few projects, the style starts to blend together.

The voiceovers are accurate but flat. They are fine for tutorials but do not carry much tone or emotion. The auto focus sometimes highlights the wrong element, and fixing that takes more time than expected.

It also requires a desktop download, which feels slightly out of step for a tool that is otherwise lightweight.

What makes it different

Kite is not about creativity. It is about getting things done. It does not replace editors or designers. It helps everyone else make something that looks clean enough to share.

It is the quiet opposite of tools like Runway that chase visual spectacle. Kite keeps things small and simple, helping teams communicate faster.

My take

Kite is a tool that knows its place, and that is what makes it useful. It is not trying to turn everyone into a designer. It is not chasing creativity or originality. It exists for one reason: to help teams show their work faster.

That focus is both its strength and its ceiling. The videos look professional but predictable, functional but never expressive. They do the job, and then they end. But maybe that is exactly what most products need right now: less theatre, more clarity.

Kite is not designed in the artistic sense. It is designed for communication. It strips away personality, but it delivers the message. And in a world of tools trying too hard to impress, there is value in one that simply helps you be understood.

In the Spotlight

Recommended watch: Kimmi K2 Thinking – the open-weights model that just outperformed GPT-5 and Claude 4.5 on humanity’s hardest benchmarks. Built as a true “thinking agent,” it can chain 200–300 tool calls, browse the web mid-reasoning, and generate everything from PhD-level math solutions to interactive data maps, all from a single prompt.

A fully open-weight model now beats GPT-5 on humanity’s hardest benchmarks. That’s a watershed moment for open source.
— Matthew Berman • ~1:40

This Week in AI

A quick roundup of stories shaping how AI and AI agents are evolving across industries:

OpenAI adds “personality presets” to GPT-5.1 – With options to switch tones from Friendly to Efficient, the latest model upgrade shifts focus from raw capability to voice and character.

Google introduces Stitch, an AI UI-design tool pitched against Figma – Stitch turns prompts or sketches into layouts and exportable code, challenging the long-standing design workflows of tools like Figma.

Microsoft releases Agent Framework Preview for developers – With APIs for agent coordination, memory and reasoning under .NET, Microsoft is placing bets on multi-agent architecture becoming a standard part of software development.

AI Out of Office

AI Fynds

A curated mix of AI tools that make work more efficient and creativity more accessible.

Krita → A professional-grade, open-source painting and illustration tool that blends classic artistry with modern AI assistance. From concept sketches to polished digital art, Krita helps creators paint, blend, and refine with precision.

Pacdora → A 3D packaging design platform that brings structure, branding, and visualization together in one place. Create dielines, customize artwork, and preview photorealistic mockups instantly, from boxes to bottles, all without leaving your browser.

AI Prompt-Based Image Editor → A browser-based AI image editor that lets you clean up, resize, and enhance visuals instantly – with smart background removal, retouching, and color optimization tools.

Closing notes

That’s it for this edition of AI Fyndings. From Hiro turning finance into structured clarity, to Kimi K2 showing the value of visible reasoning, to Kite.video reminding us that design can be simple and still say enough, this week was about AI learning to work with us and not just for us.

Thanks for reading! See you next week with more stories, tools, and ideas shaping how intelligence keeps finding its place in how we work, create, and design.

With love,
Elena Gracia
AI Marketer, Fynd

With AI, every decision is a trade-off: speed or quality, scale or control, creativity or consistency. AI Fyndings discusses what those choices mean for business, product, and design.

In Business, Anthropic brings intelligence directly into Excel.

In Product, Reducto redefines how we read and process information.

In Design, Moondream gives vision to AI that can finally see instead of just generate.

AI in Business

Anthropic Brings Claude to the Boardroom

Enterprise AI often promises big change but too often it sits on the sidelines. Anthropic’s latest move feels different. Instead of introducing another standalone tool, it embeds its AI model, Claude, directly into Microsoft Excel – the one tool almost every business runs on.

This isn’t about dashboards or new interfaces. It’s about staying in the tool you already know, with AI quietly working alongside you.

What’s interesting

I’ve always believed real innovation improves habits, not replaces them. Excel is one of those habits. It’s messy, powerful, and irreplaceable. What Anthropic is doing isn’t adding AI for novelty; it’s embedding reasoning into the daily workflows of finance, operations, and analysis teams.

Unlike general assistants that sit outside your workflow, Claude lives inside it. Imagine debugging a formula and asking, “Why is this total off?” and it walks you through the dependencies line by line. Or you’re forecasting revenue, and Claude points out an inconsistent assumption across tabs. It’s the kind of help that feels most invisible.

Compared to Microsoft’s own Copilot, which covers everything from emails to slides, Anthropic’s approach feels narrower but sharper. It’s not a layer over your work; it’s inside it.

But if you already have Copilot in Excel, why does Claude matter?

Copilot is designed for convenience. It helps automate basic commands, write formulas, or summarize sheets. Claude takes a more analytical approach. It’s built for reasoning, explanation, and financial logic. Think of it less as autocomplete for Excel and more as a business analyst that can trace logic, flag errors, and interpret data patterns.

For example,

• If Copilot can generate a formula, Claude can explain how that formula affects your model.

• If Copilot can summarize data, Claude can point out inconsistencies across tabs and suggest what to fix.

• If Copilot can fill a sheet, Claude can build a valuation model, check for assumptions, and interpret outputs.

Claude is designed around reasoning and explainability. It doesn’t just execute commands; it tells you why.

The new Excel integration also ties directly into the broader Claude ecosystem. The same model that helps you inside Excel also powers Claude for Enterprise and Claude.ai, meaning the logic and context travel across tools. You can move from analysing a spreadsheet to summarising a report or drafting a presentation without losing thread or accuracy.

Where it works well

Claude fits best for people who live in spreadsheets – analysts, finance managers, and operators. It’s ideal where data needs reasoning, not just automation.

Picture this:

• A finance team reviewing quarterly models can ask Claude to check if cost assumptions align with revenue projections.

• An analyst can ask Claude to trace a formula back through five linked sheets and show where a number breaks.

• An operator can ask it to build a scenario model, “What happens if marketing spend goes up 15%?”, and Claude will recalculate, explain the impact, and document the logic behind it.

This is what Claude is here for – helping teams understand excel better.

Where it falls short

Claude depends on data quality. If the underlying spreadsheets are inconsistent or poorly structured, even the best reasoning model can only go so far.

There’s also a trust gap to bridge. Letting an AI modify live financial models feels risky, and adoption will depend on how transparent Claude can be about every step it takes.

And for teams already deep into Microsoft’s Copilot ecosystem, adding Claude may feel like doubling up until they see the difference in reasoning depth.

What makes it different

Anthropic’s move feels more like a principle than a product launch: AI should fit into existing systems, not force you into new ones.

Instead of trying to reinvent Excel, Claude improves it. Its new “Agent Skills” for finance, like DCF valuation and coverage analysis, make it domain-aware and context-smart. While Copilot handles productivity, Claude focuses on decision quality – clarity, logic, and trust.

My take

This kind of integration is where AI starts to feel genuinely useful. It’s not about new platforms or flashy demos but more towards embedding intelligence inside the tools we already use.

If your day revolves around Excel models, forecasts, or reports, Claude doesn’t just save time. It helps you work smarter, question your own assumptions, and make data you can trust.