We had 45 minutes’ notice before testing Claude Design. No script. No benchmarks prepared. Just me, Can Hoskan, and a live recording session on the same day Anthropic pushed the update.
That’s not ideal for a fair test. But it might be the most honest one.
What We Actually Tested
The brief was deliberately simple: redesign the 02UI website to match a design reference. We used a full screenshot of the existing site, run it through Perplexity to strip out the content without any styling, then combined that with a DESIGN.md file that had the full visual direction baked in, colours, typography, spacing, the rationale behind every decision.
One prompt. No follow-up instructions. We gave the exact same brief to Lovable immediately after, same conditions, so the comparison would be fair.
Claude Design ran on Opus 4.7, which launched the same day. The server was under load. It was 5:30pm UK time, EU and US overlap hours, so we weren’t even seeing it at full speed.
What Came Back
Claude Design produced a full working redesign in under 10 minutes. The first reaction, unedited, was: “You built this in 10 minutes. You joking or what?”
What made it land wasn’t the speed. It was how much of the DESIGN.md context made it through. The output had a point of view. You could see where the references had pulled the decisions, not perfectly, not exactly as briefed, but in the direction we asked for. A senior designer could take it into Figma and develop it. That’s a different category from what AI design tools have been producing.
Lovable was faster to get started, cleaner interface, and the output was more than acceptable for five minutes of work. But it felt less shaped by what we gave it. More template, less taste.
Why the Brief Matters More Than the Tool
Here’s what the comparison actually showed: the quality of the output tracks the quality of the context.
Claude Design used the DESIGN.md file and produced something with design intent behind it. That means the effort you put into writing a proper brief, specific colours, real references, rationale for the decisions, comes back to you in the result. The tool isn’t the variable. Your input is.
That changes the question. The conversation shouldn’t be which AI design tool wins. It should be: do you know how to brief one? Because that’s where the designer’s value sits now. Not in executing pixels. In making the decisions that shape what the AI produces.
Where This Goes Next
The workflow that makes sense is: Claude Design for the first pass, then into Figma to apply a proper design system, then ship. That’s not a threat to designers. That’s a new starting point.
And here’s the thing: the barrier to entry is genuinely lower now. Someone with taste and the ability to write a proper brief can produce a result that would have taken days. That’s not designers being replaced. That’s designers cooking faster.
We’re going deeper on this in Part 2. There’s a lot more to test.
