What is human in the loop?
Human in the loop describes a system that pauses before a consequential action and requires a person to approve it. The automation prepares; the person decides.
Done well, the person sees the exact action and its exact arguments, and their approval authorises that specific proposal — not a category of future actions. Approval that is really a blanket permission is a checkbox, not a control.
The value is as much about accountability as correctness. An approved action has a name attached to it, which is what makes the difference between an incident with an explanation and an incident without one.
Approvals are bound to a digest of the exact proposal, expire rather than sitting open, and are recorded with the identity that decided.
Related terms
- Audit trail — A durable record of what a system did, who caused it, and what it touched — sufficient to reconstruct events afterwards.
- AI agent — A system that uses a language model to decide which actions to take, then takes them — as opposed to only producing text.
- Guardrails — Constraints that bound what an AI system can do, distinct from instructions that ask it to behave.
See it working rather than described.
The product demo runs one agent across three systems and stops at the approval gate.