HALF MY DESIGN SYSTEM METRICS DON'T WORK ANYMORE

Summary from the TLDR Design newsletter

Traditional design system metrics such as total component counts, downloads, MCP calls, and self-reported agent success reveal little about whether a system actually produces better work. More useful measures include conformance, accessibility, design-code parity, hard-coded values, component friction, agent adherence, cost per shippable output, and how quickly humans and agents can successfully build with the system. As AI agents become design system users, teams also need repeatable evals that give models real tasks and score their output against intended design principles, implementation rules, and visual quality.