How I use AI to deliver premium design without losing judgment
AI generates layouts, I make the calls. This is the exact workflow I use to ship design systems in days, with tokens defined before components.
Every designer asks the same question: will AI replace me? My answer after shipping design systems in production is no — but it changes where your time goes.
AI generates, you decide
My workflow starts with AI producing initial layouts from a brief. I audit, refine, and apply design judgment. The quality of the output follows the quality of the brief, not the model.
Tokens before components
For the Carga design system I used a three-layer token architecture: primitive values, semantic roles, and component tokens. They ship as Tailwind v4 CSS variables through the @theme directive. Components consume variables only — zero hardcoded colors anywhere.
AI is strong at scaffolding the token set from a spec. It's weak at naming conventions and the abstraction boundaries that keep the system coherent six months later. Those are human decisions.
What actually scales
A design system is a product, not a project. Developer experience decides whether a team adopts it. Automating the token pipeline beats writing documentation nobody reads.
The takeaway: keep AI for volume, keep yourself for judgment. See how it applies in the design work.