Agentica — the decision system for humans and AI agents
A personal design system built to prove a thesis: design decisions should survive the tools, the frameworks, and the people who made them. Zero vendor lock-in.
Context / problem
Most design systems document what was built. Few preserve why. Teams lose decisions in Figma comments, Slack threads, and tribal knowledge — so the same debates repeat, documentation drifts out of date, and the system becomes dependent on the one person who remembers the reasoning. I built Agentica to test whether that problem is solvable — not for a client, but as proof of a governance model I believe enterprise design systems are missing.
Decisions & governance
Every meaningful choice in Agentica is logged as a versioned Architecture Decision Record — context, alternatives considered, trade-offs, and the reasoning behind the final call. This isn't documentation after the fact; it's the mechanism that lets the system evolve without erasing its own history.
Agentica is also built with AI agents as a defined actor in the system — with an explicit boundary. Agents can detect inconsistencies, propose components, and draft documentation. They cannot approve, deploy, or bypass governance. The human keeps the final word, by design, not by convention.
- Decide
- Govern
- Systematize
- Execute
Architecture
One source of truth feeds four levels: foundations (tokens), semantic contracts, components, and applications. Components are built as native Web Components — framework-independent, compiled to CSS, JS, Tailwind, Angular, iOS, and Android targets. If the AI tooling around it disappeared tomorrow, the system would keep running: documented, scripted, and tested to prove it.
Result / proof
- 850+ design tokens across light and dark modes
- 17 components shipped, with more in active development
- 70+ ADRs capturing the reasoning behind every major decision
- Automated quality gates for accessibility (WCAG 2.1), visual regression, documentation, and token consistency