Agentic Design System

A design system built for two audiences: humans and AI agents. Tokens an agent can parse, components with documented intent, and rules coding tools actually respect.

All Dictionary

What it is

An agentic design system is a design system that AI tools can actually use. The components and tokens are the same; what changes is how the knowledge is structured. Tokens are machine-parseable, components document intent (when to use this, never that), and the rules live in files that agents like Claude Code, Cursor, and Codex load and respect.

A traditional design system was built for two audiences: designers and engineers. An agentic one adds a third audience, agents, and accepts that this audience cannot fill in gaps, read between the lines, or ask a teammate.

Why this matters for designers

When AI generates UI against a system it cannot read, it hallucinates: components that do not exist, invented token names, mixed variants. The output looks plausible and ships broken. The fix is not a better prompt; it is a system the agent can read. Teams that structure their systems this way get AI output that survives contact with their codebase.

How it works in practice

  1. Tokens exist in a format agents can parse (JSON, not screenshots of a Figma page).
  2. Components ship with intent documentation, not just prop tables.
  3. Rules live in context files (CLAUDE.md, skills, local markdown) that load on demand.
  4. Agents earn authority gradually, through trust levels, starting read-only.