Execution Playbook
The minimum path for launching an AI-first design system quickly.
Key takeaways
- An AI-first design system should start with an execution loop, not a perfect taxonomy, for teams moving from 0 to 1.
- The 10-day path runs from inventory and DESIGN.md through token layers, top-10 component specs, accessibility contracts, and agent instructions.
- Generated sample screens and screenshot review on days 7 and 8 expose the component and spec gaps that need fixing.
- Days 9 and 10 add CI checks for tokens, types, accessibility, and screenshots, then publish the team workflow and contribution rules.
- Definition of done: agents can find docs, are told not to invent tokens or components, and generated UI is checked at desktop and mobile.
An AI-first design system should start with an execution loop, not a perfect taxonomy. This playbook is for teams that already understand design systems and need to move from 0 to 1 quickly.
10-Day Path
| Day | Output |
|---|---|
| 1 | Inventory current components, tokens, and repeated UI failures. |
| 2 | Draft DESIGN.md with product feel and forbidden patterns. |
| 3 | Define semantic token layers and naming rules. |
| 4 | Write specs for the top 10 components. |
| 5 | Add accessibility contracts and checks. |
| 6 | Add agent instructions in AGENTS.md or equivalent. |
| 7 | Generate sample screens and run screenshot review. |
| 8 | Fix component/spec gaps exposed by generated UI. |
| 9 | Add CI checks for tokens, types, accessibility, and screenshots. |
| 10 | Publish team workflow and contribution rules. |
Minimum Definition of Done
- Agents can find component docs.
- Agents are told not to invent tokens or components.
- The system has a top-level design intent file.
- Generated UI is checked at desktop and mobile breakpoints.
- Design-system owners review contract changes.