Claude Design is just Claude Code with a UI
Written on
I’ve been trying out Claude Design for the last few days. It’s actually pretty good. I had an idea for a browser kicking around in my head, so I asked it to mock that up. It made a beautiful design on the first shot with very little prompting from me. It also mocked up a few fake websites to render inside the browser as placeholder content, which was a nice touch.
I did have to ask it to give the mock websites a different visual style from the browser chrome itself, so you could tell which parts were my browser and which parts were just example sites it had cooked up. That’s something a human designer wouldn’t have needed to be told. But once I asked, it handled it correctly and kept the two styles separate from then on, even when I asked for more changes.
Follow-up prompts work in general. I asked it to refine a few things and it followed instructions correctly, didn’t drift, didn’t redo parts I hadn’t asked it to touch. It also did a good job creating variants of a design — I asked for a compact version of a tab bar and it came up with a decent looking one that only needed a couple rounds of feedback to get right. So as a tool, it’s okay.
But the more I used it the more I realized: it’s just writing jsx and css behind the scenes and rendering the result for you. That’s it. You could do the same thing with regular Claude Code. Ask it to spin up a React project, ask it to come up with designs, and then hook up Claude in Chrome so it can screenshot the result and verify what it made. That’s all Claude Design seems to be doing under the hood, just with a Figma-shaped UI around it.
Which is why I’m a little confused about the billing. Design has its own usage bar, separate from my regular agent usage, and it burns down really fast. I assume the separate bar is an introductory thing — Anthropic giving you a free taste so you can try it without nuking your regular usage. But once you run out, they want you to pay extra for it. Which I think is how they’re gauging interest. Nothing says “I want this” louder than putting your money on it.
There is the sketch-a-design-and-the-AI-follows-it feature, which is cool in theory. But I honestly don’t know how closely it follows the sketch versus just looking at the rough vibe. And as far as I can tell that’s just feeding the image into Claude, which you could also do by drawing the sketch yourself and dropping the image into Claude Code.
The other thing that bugs me is that there’s no way to “commit” your changes. The whole thing feels ephemeral. If I were doing this in Claude Code locally like I described, I’d ask the agent to commit whenever I hit a state I was happy with, so if it broke something on the next round I could roll back. With Claude Design it feels like the agent could trash something and I’d have no way to recover. You can manually download the files but that’s a hassle, and we invented version control for a reason. I don’t see any kind of versioning here at all. You can ask the agent to make a new file, sure, but I don’t see anything stopping the agent from touching files you didn’t ask it to touch either.
There’s also the collab side: comments on designs, sharing, that kind of thing. As a solo dev that’s just not a concern for me at all. And on a team I still don’t think it’s a deal breaker, because we already share screenshots and looms and Slack threads when something needs eyes on it. The only time I personally end up in Figma is when a designer wants feedback or is handing something off.
So I’ll keep using Claude Design while it’s free-ish for me. But if I start hitting the usage limits — and the way my usage is climbing, I might — I’ll just switch to spinning up a React project in Claude Code and pointing Claude in Chrome at it. I wouldn’t pay extra for the wrapper.