AI Does Not Read Your Design System the Way You Do
1. Components alone do not convey the reasons behind decisions
AI can build screens quickly, but it does not know the usage context the team has defined.
- Components are the result of countless decisions compressed into one artifact.
- When to use them, which combinations to avoid, and why certain variants were discarded are hard to understand from appearance alone.
- People fill in the blanks by talking with colleagues or recalling experience.
- If AI has no records to reference, it can produce UI that looks correct but differs from the team’s standards.
2. From the number of components to “decision coverage”
The original article proposes Decision Coverage as a new metric for the AI era.
- Component coverage looks at how much of the UI is implemented as a library.
- Decision coverage looks at how much of the judgment behind that UI is recorded in a machine-readable way.
- It includes usage conditions, prohibited cases, exceptions, and the reasons for them.
- Even teams with the same components can get different AI results depending on how complete these records are.
3. Connect records, resolve conflicts, and verify
Providing more materials alone is not enough.
- Connect component rules with user journeys and service context.
- If design files, code, and documents differ, decide which standard to follow.
- Keep the source of the answer so the basis for the judgment can be traced.
- Evaluation only becomes meaningful when agreed-upon standards are compared with AI results.
