Anthropic shipped Claude Design on Friday.
Every launch-day publication called it a Figma killer.
After six hours inside it on launch day — with my production codebase as the input and Claude Code on the other end — I think that framing misses what actually shipped.
Claude Design is not a design tool.
It is the missing front-end of a four-stage loop that already existed in pieces across the Claude product surface:
- Idea capture — prompt, screenshot, codebase pointer
- Codebase-aware design — your colors, typography, components, extracted automatically
- Claude Code handoff — local CLI agent or Claude Code Web
- Shipped product — inside the agent workflow you already use
None of the stages is new. What is new is that they live behind a single product URL.
The only teams who can see this clearly are teams already running Claude Code.
I pointed it at openLEO, my productized OpenClaw platform. The color palette and typography lifted cleanly from the codebase. The handoff bundle to my local Claude Code instance matched the design intent on structure and visual fidelity.
Where it was thinner than I wanted: state specifications and animation patterns. Claude Code filled those with reasonable defaults. For now.
The most useful page Anthropic shipped is not the launch announcement. It is the four-bullet "Known limitations" section in the docs. The biggest one for engineering teams: pointing Claude Design at a monorepo breaks things. Link subdirectories instead.
Six hours is not a verdict. Research preview features will change. Token economics at team scale are still unknown.
But the loop is real. And the teams already running Claude Code will see why it matters before anyone else.
Full breakdown with five documented limitations and the handoff test: [MEDIUM URL]
What does your own design-to-ship loop look like today? Figma MCP stitched to Claude Code? All-in on Claude Design from day one? Somewhere in between?
Top comments (0)