A trust tier is the autonomy level an agent has earned: watch, suggest, draft, or act with undo. Tiers are per agent, backed by a visible track record, capped by the owner, and revoked by corrections.
Trust tiers turn the scariest question in agentic AI, 'what is it allowed to do?', into an inspectable object: this agent, this tier, earned on this record, capped here.
The mechanics live at falkster.ai/answers/approval-tiers; the philosophy is one line: trust is earned, not given.