Trust Levels

A five-step scale (Observer, Advisor, Junior, Senior, Autonomous) that defines what an AI agent is allowed to do, and what it has to prove before it gets more authority.

What it is

Trust levels are a ladder of permissions for AI agents. Instead of deciding “do we let AI touch our design system: yes or no?”, you give each agent a level: Observer (read-only), Advisor (suggests), Junior (acts with review), Senior (acts with spot checks), Autonomous (acts alone in a defined scope), plus a clear bar it must clear to be promoted.

Why this matters for designers

Most AI failures in design system work are authority failures, not capability failures: an agent was allowed to change something it should only have been allowed to flag. Trust levels turn a vague anxiety (“can we trust it?”) into an operational question (“what level is this agent at, and what does the evidence say about promoting it?”).

Try it

The fastest way to understand trust levels is to feel the difference. Same agent, same finding. Switch the level and watch the blast radius change:

Trust levels playground

Same agent. Same finding. Five levels of authority.

AID's audit agent just found drift: Button.tsx uses the hardcoded hex #3B82F6 in 3 places, and the token color.action.primary exists for exactly this. What happens next depends entirely on the agent's trust level. Pick one.

Reads and reports. Touches nothing.
Agent

Scanned 87 components in 4 minutes. Report ready.

Agent

Weekly audit report:

audit-report.md
3 hardcoded hex values in Button.tsx
Matching token exists: color.action.primary
No other violations found this week
You

You read the report. That is the whole loop, and that is the point. You are learning what this agent reliably catches and what it misses.

Same agent, same finding, five different blast radii. The level, not the model, decides what happens next.

How it works in practice

  1. Every agent starts at Level 0 (Observer): it reads and reports, nothing else.
  2. Promotion is based on its track record: accuracy of findings, false-positive rate.
  3. Each level is scoped: an agent can be Senior on token files and Observer everywhere else.
  4. Demotion is always on the table, and the kill switch is not optional.