What happens when AI can see every designer’s PR?

Main point: PR review catches mistakes in one piece of work but misses patterns across many. An AI that can see every designer's PR can spot those patterns and help the whole design group work more coherently.

Individual review

Before raising a PR in Unplanned Studio, designers run an AI audit against a checklist (tokens, accessibility, TypeScript strictness). Human reviewers then see work that has already passed the mechanical checks. Useful, but table stakes.

Group review

Looking across all PRs, an agent can answer questions no single reviewer can:- Which conventions are broken most? If hardcoded hex values are in 40% of PRs, that's a discoverability problem. Fix the docs or the skill, not the designer.- Which components keep being rebuilt? Three home-made filter toolbars in a month means the library is missing one. (Related: [Why "throw-away code" is the wrong frame for agentic design prototyping])- Where do new designers struggle? First PRs show the gaps in onboarding.- Which patterns are becoming a standard on their own? Those are ready to promote.

GitHub ties it together

A PR links to the Issue (why), the Project (where it sits) and comments (human judgment). With all of that, the AI can surface what a retro normally would. (Related: [GitHub Issues for Design Teams])

Review as monitoring, not surveillance

The point isn't to catch people out. It's to see where the design system works and where it causes friction. Human judgment on design stays.

What it needs

A shared repository, and well-defined skills and checklists for your system. Generic AI review is shallow. The return is an organisation that learns from its own output.

Related: [The Design Organisation as a Codebase], [The Informed Branch], and A first-principles approach to design in the age of AI.