Had a fun weekend building Merisole ☀️ Designed with Google Stitch.

I'm a senior product designer working on e-commerce sites every day, so my real question going into this challenge was practical: how easily does Stitch slot into my actual workflow? Could I use it to spin up a polished, client-ready option in a single sitting or even take it all the way to a shippable site? To test that, I asked AI to create a fictional footwear brand, Merisole, and treated it like a real client project from brief to launch.

How Stitch fit into the workflow Before generating anything, I wrote a custom design.md. Type scale, color tokens, spacing rhythm, the editorial tone I wanted Merisole to carry and uploaded it into Stitch as a design system. That single step is what made the rest of the build feel coherent. Every screen Stitch streamed onto the canvas already spoke the brand's language, so I wasn't fighting drift between sections. This is exactly the part of my normal e-com workflow that usually eats hours! Getting AI tools to respect a brand spec was something that Stitch handled cleanly. By the way, the stream on canvas feature was one of the coolest interactions I've seen so far! It was like having a designer design on your screen remotely through AnyDesk!

From there I worked mostly through in-place AI edits: clicking a section, telling Stitch "to add it to another variation," or "add a hover animation on the cards," and getting changes in seconds. The streaming generations meant I was reacting to a real interface taking shape instead of waiting for a static result.

The motion work is where Stitch surprised me most. Native hover states, shader backgrounds, and scroll-linked animation. All authored inside the tool, no handoff to a separate motion app, no code wrestling. Then I exported the build out to Netlify in one step.

New features I leaned on

Custom design.md - design system applied across the whole project (kept every screen on-brand from the first generation)

Streaming generations to canvas (the whole first pass)

In-place AI edits via prompts + point-and-click (most of the polish phase)

Native motion, hover states, and shaders on HTML canvas

Netlify export (final deploy)

Feedback on the platform Coming at this from a daily e-com design seat: Stitch genuinely shortens the distance between brief and client-ready. The design.md - design system flow is the underrated hero, uploading a brand spec once and having Stitch honor it across every streamed generation is the closest I've felt to "AI that respects a brand." The streaming canvas is amazing to say the least! It collapses the "describe - wait - judge - re-prompt" loop into something closer to direct manipulation. In-place edits felt the most like real design work. I stopped writing long prompts and just pointed at things. Two things I'd love next: An option to tweak elements without having to prompt it. And a way to save in-place edit "recipes" so I can reapply a vibe across sections without re-describing it.

Stitch didn't feel like an AI tool I was wrangling. It felt like a canvas that happened to understand what I meant. Merisole came together in a single sitting because of that, and I'm walking away convinced Stitch has a real place in how I'll pitch and build e-commerce work going forward.

Check it out:

Live site (really fun to click around): https://merisole.netlify.app

Stitch project: https://stitch.withgoogle.com/projects/834230787052196219

Amazing!

Designed the UI/UX for an AI assistant web app where the challenge wasn't the AI - it was taming 15+ customization settings without overwhelming users. Used progressive disclosure: new users see 3 settings by default, power users can expand the full suite without the interface ever feeling different. The result is a product that scales with the user's comfort level instead of front-loading complexity.

Check it out

The progressive disclosure approach for the AI settings is brilliant UX design! Starting with 3 settings and expanding for power users is exactly the right approach.