The design system is becoming an AI instruction layer
Coinbase’s design systems team reports that Code Connect improved agent context, reduced token costs by 22.5%, and cut task time by about 22%.
Reported average reduction in agent token costs using Code Connect.
The Story
Coinbase’s design systems team tested what happens when coding agents can see the relationship between Figma components and their production counterparts. The result is less guesswork, better adherence, and a more legible path from canvas to code. ([figma.com](https://www.figma.com/blog/how-coinbase-used-code-connect-to-shrink-token-costs/?utm_source=openai))
The useful shift is conceptual: a design system is no longer just a library for humans. It is also a set of instructions, constraints, and examples that can steer machines toward the intended product language. ([figma.com](https://www.figma.com/blog/how-coinbase-used-code-connect-to-shrink-token-costs/?utm_source=openai))
For Studio Wensday, this strengthens the case for documenting not only what a component looks like, but why it exists, when it should be used, and what “good” means in context.
Studio Wensday Angle
Design systems as operational knowledge for agents.