Design systems were built to help humans agree on what a button looks like

Design systems were built to help humans agree on what a button looks like. That contract is being renegotiated.

AI agents don't read documentation the way designers do. They parse context. They assemble interfaces from parameters, not prescribed patterns. The system you built for team coordination is becoming runtime input for automated UI generation.

I've been thinking about this from the bank side. The design system we built assumed a designer was always in the loop, interpreting, adapting, making judgment calls. If an agent now queries that system and generates screens without human review, the assumptions baked into every component become literal rules. Not guidelines.

The quality bar shifts. Not just "did the designer follow the system?" but "does the system itself hold up under inference?"

Does this make the design system architect more important, or just another constraint that slows down automation?