What's this about?
Anthropic shipped a research preview of `/design`, a skill that brings Claude Design's artboard workflow into Claude Code — both the CLI and the Desktop app — built on artifacts. The suggested usage is to design before building: run `/design a few options for {feature}`, get several editable artboards for your interface, pick the one you like, adjust it, then have Claude implement it.
What exactly is an "artboard"?
An artboard is a design-tool concept — a fixed canvas representing one screen or one state of an interface, which you can lay several of side by side to compare variations. It's how designers work in Figma or Sketch: not one document, but a wall of alternatives you evaluate together. Bringing that into a terminal-based coding tool is unusual, because a CLI has historically been the least visual environment in software. The point is that choosing a direction is a visual judgment, and until now the agent could only describe options in prose or jump straight to writing code.
Why is this worth noticing?
- It closes a real gap in the agent loop. Coding agents have been strong at implementation and weak at the step before it — deciding what to build — where several rough options beat one confident answer.
- It builds on artifacts rather than a new surface, meaning the design output is a live editable thing rather than a picture, and the implementation step already has the exact source.
- It arrives alongside a related observation from Thariq the same day: recent procedural-generation, video-editing and 3D game demos have him updating toward LLM coding models being better at creative work than diffusion models — the argument being that the creative parts we care about live at the level of intelligence rather than technique.