I designed an app in Claude before building it. Claude Code shipped it in a day.

My first time using Claude Design to plan and prototype an app before handing it to Claude Code. A cross-channel client reporting tool, designed and built in under a day — and what it means for my use of Lovable.

This week I did something I'd never done before: I designed an app before I built a line of it. Not in Figma, not on a whiteboard — in Claude Design, Anthropic's prompt-to-prototype tool, where you describe what you want and it renders a real, interactive interface on a canvas. Then I handed that off to Claude Code and let it build the actual thing. The whole app was live in under a day.

I went in genuinely unsure how close this would feel to working in a tool like Lovable. The answer turned out to be: closer than I expected, with one honest caveat.

What I actually built

The thing I wanted has been on my wish list for years, and no single reporting tool has ever given it to me cleanly. For a client I look after, I built a cross-channel reporting app that pulls together the pieces that normally live in five different back ends: marketplace sales, web analytics, and ad spend from across the major ad platforms and a couple of regional sales channels, all merged into one order report that refreshes itself through scheduled routines.

The view I really wanted was simple to describe and annoyingly hard to get anywhere else: yesterday's sales across every channel, with all the costs pulled in alongside, and return on ad spend broken down by country and product category. On top of that, the part I love most — lead-generation funnels and customer acquisition cost tracked against what a customer is actually worth over their lifetime. That lifetime-value-against-acquisition-cost picture is the number that tells you whether you're really making money, and it's almost never sitting in one dashboard.

So I built it into a single standalone app. Something I can open every morning to track client work, and clean enough to share with clients and partners so they get a real overview instead of a screenshot.

Lovable still wins on polish and logic. But I can design it in Claude, have Claude Code build it, and host it myself — no token meter, no walled garden.

Design-first changed how the build went

Here's what surprised me about doing the design first. By the time I handed anything to Claude Code, the hard decisions were already made and visible. Layout, hierarchy, what goes on the daily overview versus what's a click away — I'd seen it rendered and reacted to it, instead of discovering it halfway through a build. Anthropic is explicit that Claude Design is meant to export into Claude Code, and that handoff is the whole point: the design tool gives you fast visual iteration, the coding agent gives you a real repo, real data connections, and tests.

It's the same lesson I keep relearning with these tools — the quality of what comes out depends on the context and examples you put in. Reviewers describe Claude Design as a genuinely new way to brainstorm visually with AI, and that matched my experience: feed it clear references and a precise description of the job, and it gives you something you'd actually ship. Feed it mush, and you get mush. The leverage is in the brief, which is exactly the argument I made about domain expertise being the real edge.

Where this leaves Lovable for me

I'm not abandoning Lovable. It's still more polished out of the box, and its logic and styling defaults are further along — I've built real multilingual setups in it and it earned its keep. But the calculus is shifting. With the Claude Design plus Claude Code route, I can build tools that look great and host them myself, without buying tokens inside someone else's system or living within their hosting. For a builder who already has the context and the judgement, that independence is worth a lot.

This app is also the companion to the discipline I wrote about separately this week — once you've merged four sales channels into one number, the next job is proving the number is right. The build is the easy half; the trustworthy data underneath is the real work, and a dashboard that leaves things out is worse than no dashboard. Designing it beautifully just means more people will actually look at it — which raises the stakes on it being correct.

A year ago this would have been a multi-week project with a designer, a developer, and a hosting headache. This week it was an afternoon of design, an afternoon of build, and it's live and kicking. I'll be reaching for this combination a lot more.

Sources & further reading

External — Claude Help Center, Get started with Claude Design; Builder.io, Claude Design review: an innovative way to brainstorm with AI; MindStudio, Claude Design vs Claude Code: which should you use for UI and prototypes?

Related postsI merged four sales channels into one number. Then I proved it was right.; Beyond the remix: a multilingual setup for Lovable that actually scales; The "vibe coding" breakthrough; Your analytics dashboard is lying by omission; The death of generic AI.

Subscribe to Remco Livain

Don’t miss out on the latest issues. Sign up now to get access to the library of members-only issues.
jamie@example.com
Subscribe
Work with me →×