DesignSpark AI
An agent that generates and maintains tokens and living documentation for Design Systems.
Visit product
The context
Tokens drifted between web and mobile, and documentation went stale the day after every release. The cost wasn’t visual — it was the distrust it seeded in the system.
The real problem
The tension was autonomy versus integrity: designers wanted to version components without waiting on a dev, but without governance that becomes token chaos.
My role
- Defined the single-source-of-truth architecture for tokens.
- Designed the agent flow inside component creation/update.
- Validated integrity and rollback with UI engineers.
The critical decisions
A changelog designers can read, not just devs
The change log had to be understood by the people who design. Writing for the designer, not for git, is what got governance adopted.
Governance in the component CI/CD
Baking rules into the pipeline instead of relying on manual discipline kept consistency from depending on goodwill.
The AI layer
- Prompt templates with real examples prevented generic output.
- Every generated change passed through an auditable changelog and safe rollback.
- The designer approved before any merge — the AI proposed, the human decided.
Outcomes
- Fewer token divergences between web and mobile.
- Per-release documentation became significantly faster.
- Squads started adopting PRs with an automatic changelog.
What I’d do differently
I underestimated how much trust in the agent hinged on changelog readability. Today I’d start there, not with token generation.