Anthropic is extending Claude Code, the command-line version of the AI assistant, with a design step. With the /design command, interface drafts can be generated directly in the terminal, even before the first line of implementation code is written. The results appear as artboards, i.e. delimited design canvases as known from classic design tools.

The key difference from a separate design tool lies in the context: Claude reads the existing codebase and aligns the drafts with the UI style already in use. Instead of a suggestion on a blank canvas, a draft emerges that picks up the patterns established in the project.

Why the place within the tool matters

In many projects, there is a media break between design and implementation. A design is created in a graphics tool, exported, commented on, approved and then translated into components by developers. Details are lost at every transition: spacing, states of controls, behaviour at small viewports.

When the draft is created where the code lives, this workflow shifts. The suggestion already knows the building blocks that exist in the project and does not have to be traced back to them afterwards. This reduces precisely those breakpoints that cause rework in practice.

What this means for prototypes

For coordination with clients, speed is above all what matters. A draft that is on the table in minutes rather than days changes the nature of the discussion: people talk about a concrete variant instead of a description. And because variants cost little, several directions can be shown side by side instead of committing to one early on.

At the same time, the character of these results remains that of a draft. An AI-generated artboard does not replace engagement with the target audience, information architecture or brand management. It shortens the path to the first visible version — the substantive decision about which structure is the right one remains with those involved.

Classification for project practice

The announcement is part of a broader movement: tools that previously covered separate project phases are growing together. Claude Code was initially an assistant for code in the terminal; with the design command, it intervenes in a phase that traditionally comes before development.

For collaboration in teams, this raises questions that are not solved technically:

  • Approvals: When drafts are created by the second, a clear rule is needed for which variant is binding.
  • Design system: The cleaner the existing component library, the more usable the generated suggestion. Untidy code leads to untidy drafts.
  • Roles: Who checks accessibility, contrasts and interaction logic when a separate design step no longer takes place?

These points do not argue against the approach. But they show that the benefit does not come from the tool alone, but from the process into which it is embedded.

How to start sensibly

A realistic entry point is a clearly defined use case: a new detail page, a form, a dashboard section. There you can observe how well the generated drafts match the existing style and where manual rework is needed.

It is equally sensible to check your own component base beforehand. Clearly named building blocks, consistent spacing and documented states are the prerequisite for a tool to be able to correctly recognise the existing style at all. This work pays off regardless of whether AI support is used.

For web and software projects, this means: the boundary between the design and development phases becomes more permeable, and the first visible version moves significantly closer to the start of the project. The greatest benefit is drawn by teams whose codebase and design system are already tidy — because that is precisely what the quality of the suggestions feeds on. Anyone planning to use such tools should therefore think less about licences and more about their own component structure and approval paths.

Sources