Human Judgment and the One Gate

Human-in-the-loop approval means a person resolves an action before it executes. It is appropriate where consequences, uncertainty, or required authority call for human judgment. It is one way to resolve Figaro’s gate.

One boundary for consequential action

Figaro’s authority model keeps exactly one resolution boundary. A human, deterministic policy, or watchdog can be the decider only within an explicit grant. Proposing an action and holding permission to resolve it are separate powers. Neither job title nor access to an AI tool creates authority.

Give the decider a complete decision

A useful proposal contains the exact change, accountable owner, reasoning, dated evidence, alternatives, consequences, and expected outcome. It states how and when the result will be measured. Missing evidence is a gap to resolve, not a zero to fill in.

Move human attention toward the decisions that need it

Humans can grant routine action classes bounded authority and review samples of the resulting record. Novel, irreversible, or high-exposure work may stay human-resolved. Urgency never creates permission. The record must show who resolved the gate and under which authority.

Money movement is denied by default. A client may grant bounded exceptions only inside the applicable law, provider, and platform limits. There is no general permission to spend.

The gate stays as autonomy grows

Delegation happens at the gate. It does not create a second route around it. An agent’s good results cannot widen its own limits. Read the autonomy progression for how evidence and grants fit together.

Current availability

The existing native operator system uses human approval for consequential writes. Policy and watchdog resolution are part of the Powered by Figaro rollout, with coverage verified per installation. The product map distinguishes those layers.