I Tested Claude Design: Here's What Happened to My UI

Anthropic dropped Claude Design this week, and the internet immediately went into "is this the end of [insert creative profession here]" mode. This time it's designers in the crosshairs—specifically UI designers and maybe even Figma itself.

Developer OrcDev decided to actually test the thing instead of speculating, spending a few hours pushing Claude Design through its paces on his GitHub Creature project (a fun side project that generates trading cards from your GitHub stats). What he found is more interesting than either the hype or the dismissal would suggest.

What Claude Design Actually Does

The pitch is straightforward: sketch out a rough UI, feed it to Claude Design, and get back a full design system with components, color palettes, typography scales, spacing grids—everything you'd expect from a proper design foundation. OrcDev's test generated design systems with type hierarchies, rarity palettes for his card game, dark mode surfaces, motion specs, and even accessibility guidelines.

"You can see here that we have everything that is including animations and everything that is moving around inside of the project," he notes while touring the generated design system. The tool created subtypes for fire, water, earth, and air elements based on his card types, which he actually appreciated. It understood context.

For the actual design test, OrcDev sketched a basic three-section layout: navbar, hero section with cards, and a features area. One prompt later—after about 10 minutes of processing—Claude Design returned a complete landing page. The navbar looked okay. The hero section had "really nice animation." The features section was "definitely better than what I already have."

So... success?

Where It Falls Apart

Not quite. The devil's in the details, and Claude Design has a lot of devils.

First: the tool generated fake cards instead of using actual ones from OrcDev's GitHub integration. It had access to the real data but decided to improvise anyway. Then there's mobile responsiveness—or the complete lack of it. "Everything is broken on mobile," OrcDev observes. Features collapsed into single cards, horizontal scrollbars appeared, the whole thing fell apart on smaller screens.

The annotation feature, where you can circle elements and ask for changes, literally broke while he was testing it. Elements moved around randomly while scrolling. "This looks like human eye didn't even catch up what's even happening on this Claude Design," he says, and honestly, it's hard to argue.

Claude Design itself doesn't even look good at smaller resolutions. The irony of a design tool with design problems isn't lost here.

The Anthropic Pattern

What's fascinating is how OrcDev contextualizes this within Anthropic's broader strategy. He sees Claude Design as part of a ship-fast, iterate-maybe approach:

"I think this is just another like 3-day project from Anthropic. They just vibe coded this whole thing and they call it a product. Of course, they are home like 400 billion company. They can afford anything and they can just ship like mad like 10 products a week."

No dark mode in a design tool launching in 2024. PDF export and "send to Canva" options, but no Figma integration despite Figma supposedly being the threatened party. These aren't oversights—they're choices that come from prioritizing launch over completeness.

This raises an interesting tension: should startups and big tech companies approach shipping differently? OrcDev suggests yes: "When you're shipping something, don't wait for perfection, but chase it. But don't chase it like Anthropic. They're not even chasing it."

It's a distinction worth sitting with. Moving fast makes sense when you're scrappy. But when you're valued at $400 billion, is "vibe coding" still a defensible strategy, or does it become carelessness with a budget?

The Taste Problem

Here's where OrcDev lands, and it's the most interesting part of his argument:

"AI doesn't have a taste. And it has all the best practices. It can get the right color for your type of niche, business, or whatever it is, but it can never have a taste like a senior design engineer. And it doesn't have eyes as well."

This isn't a Luddite dismissal. It's a specific claim about what design actually is. Claude Design can follow rules, apply best practices, generate technically correct color palettes and spacing systems. What it can't do—at least not yet—is make the kind of aesthetic judgment calls that separate good design from great design.

Or as OrcDev puts it more bluntly: you can't ship those buttons. You just can't. And knowing why you can't requires taste, not just technical competency.

What This Actually Means

So is Figma stock dropping because of Claude Design? Probably not, considering Claude Design's export options pointedly exclude Figma integration. The tools serve different purposes. Figma is for collaboration, iteration, and the kind of detailed design work that happens after you've moved past the "okay-ish" stage.

Claude Design seems more useful for developers who need something better than their current placeholder UI but don't have design resources. It's a productivity tool, not a replacement tool. The fact that OrcDev needed to caveat nearly every positive observation with "but I'd need to iterate" or "this isn't quite right" tells you where the ceiling is.

The real question isn't whether AI can generate designs—clearly it can. It's whether generating designs is the same thing as designing. OrcDev spent 2-3 hours testing and got something that was "okay-ish" and "not perfect" but "definitely better" than what he had.

A senior designer might have spent the same time and produced something actually shippable. And they would have known, without testing, that those buttons were unacceptable.

That knowledge—that taste—is still the thing that matters. At least for now.

— Zara Chen, Tech & Politics Correspondent