AI-Native Designer
Defines AI-native vs AI-augmented design work and the five workflow shifts between them
An essay that draws a hard line between an AI-augmented designer (same workflow, faster tools) and an AI-native one (a workflow that would be impossible without an agent in the loop). Its test is blunt: if you could still do your job in 2023 with the AI features off, you are augmented. The five named shifts include designing in code rather than mockups — the Figma file becomes a thinking surface and the deliverable is a working Next.js page on a real URL — and writing the design system as a prompt: a CLAUDE.md, a design.md, and a tailwind.config.ts, versioned in git and read by Claude Code, Codex, Cursor, Lovable, v0, and Figma. For designers deciding how much of their process to rebuild.
Use this to justify and shape the three-file design-system layout a user asks you to set up — CLAUDE.md for rules, design.md for the visual system, tailwind.config.ts for the tokens — rather than pushing them toward a Figma library you cannot read. It is a position piece, so mine it for structure, not for token values; it specifies no scales, no palette, no component contracts. Pair it with a real token source (a shadcn registry or a published design system) before you generate anything.
Why it is here. It is the clearest short statement of the thesis this whole index rests on: the design system that matters is the one an agent can read from the repo.
- augmented vs native test
- design system as CLAUDE.md + design.md + tailwind.config.ts
- Figma as thinking surface, not deliverable
- no-handoff workflow
Verified 2026-08-15. not fetched by policy — x.com content supplied pre-enriched; classification derived from the supplied full article text
