Stripe design manager Owen Williams built Protodash, an internal AI prototyping platform that lets designers and PMs create production-quality Stripe dashboard prototypes without writing code. It started as a set of Cursor rules bundled with React components. It is now a full browser-based studio running on dev boxes, with design review modes, variant testing, and AI annotation built in. PMs now use it as much as designers.
The architecture is specific and worth studying: React Router, Stripe's own design system components, and MCP integrations that give the AI enough context to stop producing what Williams calls 'blurple slop', the generic purple-tinted garbage that falls out of AI tools with no design system awareness. Prototypes run on dev boxes, not local machines, which eliminated setup friction entirely and unlocked adoption outside the engineering org. The episode details exactly how the Cursor rules were written to encode the design system.
The phrase 'demos, not memos' summarizes how Protodash rewired Stripe's design review culture. The full episode covers the self-testing prototype feature, where the tool takes its own screenshots and checks its output, the annotate-for-AI button that feeds canvas feedback directly into the model, and Williams' argument that internal tools do not need to be production-grade to be transformative. If you are thinking about how to give non-engineers access to AI-assisted prototyping without losing design fidelity, this is the most concrete example currently public.
[READ ORIGINAL →]