The AI-ready design system checklist

Each point you miss is a place where an agent guesses. Start with the lowest: a part marked No is one the agent has to invent every time it builds.

A checklist is easier to trust once you have seen it used. Here it is on Microsoft Fluent 2 (React v9), one of the public systems in the comparison. Each answer rests on a page of Fluent’s own, listed under Sources, and the four it rates match the comparison, which leaves typed props out.

1. Tokens as data, Yes for 2 points. Fluent publishes its design tokens as a package, @fluentui/react-theme. They are TypeScript objects, not a JSON file, which the checklist’s Yes allows because it counts a tokens package. It is the one judgment call in the example.
2. Component APIs and typed props, Yes for 2 points. The @fluentui/react-components package ships its type declarations in dist/index.d.ts. Typed props are the part the comparison leaves out, and the checklist asks.
3. Agent access, No for 0 points. A Fluent maintainer wrote on 2026-02-25 that Microsoft’s own Fluent MCP server has not been released outside Microsoft. Community servers exist, but the checklist asks for one made by your own team.
4. Readable docs, Yes for 2 points. Fluent’s llms.txt links a plain-text file for each docs page and component.
5. Explicit guidance, Partly for 1 point. Fluent’s repository has an AGENTS.md, but it is written for people contributing to Fluent itself. That is what the checklist’s Partly describes, instructions only for people who work on the design system’s own code.

The total is 2 + 2 + 0 + 2 + 1 = 7 of 10, which the checklist above calls Nearly AI-ready. The comparison shows the same four ratings: Yes for design tokens, No for agent access, Yes for docs and Partly for guidance. One caution, though. A score describes what agents can find, not how they behave. The test set of prompts measures that.

What to do at each band of the score

• 10 points, AI-ready. Keep the checks running as the system grows, and run your test set again whenever the design system, its guidance or the model changes.
• 7 to 9 points, Nearly AI-ready. Fix the Partly and No answers, lowest first. For a team on Fluent that means writing its own guidance for agents and bringing its own agent setup, since no guidance for people building with Fluent and no official server were found.
• 4 to 6 points, Partly AI-ready. Start with the parts agents read before they can ask, which means design tokens as data and docs an agent can read. Add a design system MCP server when agents still pick the wrong parts.
• 0 to 3 points, Not AI-ready yet. Agents are working from what they learned elsewhere. Begin with design tokens as a file and a short AGENTS.md for your design system, then run a bounded pilot on one surface to see what changes.

See how the public systems measure up in the best AI design systems, compared, or read what is an AI-ready design system first.