When a colleague spins up a working prototype from a user story in under half an hour — no Figma file, no handoff ticket, no designer in the loop — the reaction is not “this is impressive.” The question that follows is: which parts of how we work were built because this used to be expensive?
That’s the question Claude Design keeps asking me.
The pattern that keeps repeating
Claude Design shipped in research preview on April 17, 2026, and the predictable takes followed: Figma’s stock dropped 7 percent, some commentators argued the rough edges meant it wasn’t ready, others declared it would replace every designer. All of that missed the structural point.
Anthropic has now shipped the same architecture three times. Claude Code turned a natural-language description of software into working code, with no engineer queue in the middle. Cowork did the same for documents and analyses. Claude Design extends the pattern to visual artifacts — pitch decks, interactive prototypes, 3D configurators, full design systems extracted from a codebase in about 20 minutes. In each case: intent goes in as plain language, an artifact comes out, and the specialist queue in the middle becomes optional.
When Code launched, the question was what developers would do with the hours freed up. Now it’s design’s turn.
What the cost collapse actually means
For roughly two decades, the prototype — a realistic, interactive mock of a feature before anyone built it — was a discrete, expensive step owned by a specialist. It existed as a separate artifact because producing one used to take designer-weeks. The economics created the workflow: separate roles, separate tools, translation losses at every handoff, a QA pass to verify that what engineering shipped matched what the designer intended.
What’s collapsed isn’t the need for good design judgment. It’s the cost of the object that judgment used to produce. When ten directions in an afternoon becomes routine, rationing disappears. You stop asking “should we prototype one direction or two?” and start asking “which of these ten directions is actually right?”
That’s craft moving upstream. Less time making mockups; more time deciding which ideas are worth pursuing and why.
Four prompts to act on this
Here are the four prompts worth running when orienting a design team to this shift.
Design System Extractor. Before anything else, build a portable design-system spec — a file that captures your brand’s color tokens, type scale, and component rules — from your existing codebase and brand artifacts. This becomes the opening context for every Claude Design session. It’s the difference between generic output and output that looks like it came from your team.
PM Prototype Sprint. Feed in user stories and acceptance criteria. This prompt produces a Claude Design brief that generates a working, multi-state prototype — empty state, error state, loading state, happy path, high-volume edge cases — all pre-specified. The prototype goes in the Jira ticket instead of the spec document. What used to take designer-days takes about 30 minutes.
Org Role Audit. This is the one to run first. It looks at which roles, handoffs, and rituals on a team exist because prototyping was expensive — not because the work genuinely requires them. The five-person mockup team, the QA pass verifying that design intent survived translation, the PM-to-designer-to-engineer chain. Some of that is still necessary. Some of it was scaffolding around a cost that’s now gone. The output is a structured breakdown of what stays, what moves upstream, and what can be retired — useful to have in hand before the leadership conversation, not after.
Figma Migration Decision Tree. Figma still wins on collaborative editing and maintaining design systems across large organizations. Claude Design wins on speed and on going straight from spec to interactive artifact without a proprietary file in the middle. This prompt helps teams decide where each tool fits rather than treating it as a binary replacement question.
The key architectural question
Google’s Stitch made a different architectural bet than Anthropic. Stitch introduced DESIGN.md — a plain-text file describing design tokens, type scale, and component rules that the AI reads before generating anything — as an open, shared standard. Anthropic’s bet is integration: extract the design system from the codebase automatically, then hand off directly to Claude Code for production. Both approaches agree on the same underlying shift: code and plain text have beaten proprietary design file formats as the durable medium for AI-generated artifacts.
The interesting question isn’t which approach wins. It’s what design teams will be doing once the economics settle. My bet is that the most valuable designers — like the most valuable developers after Claude Code — will be the ones who got clearest about what judgment, taste, and direction actually mean when the cost of the artifact they express those things through approaches zero.
Text summarized and optimized using Anthropic’s models and reviewed by a human.