Human-in-the-loop

Human-in-the-loop is a system design in which an automated process pauses at defined points and waits for a person to approve, reject or amend before continuing, rather than running to completion unsupervised.

The important word is design. Human-in-the-loop describes where the pauses are built into the system, not whether somebody happens to be watching. A process that a person could interrupt is not human-in-the-loop; a process that cannot proceed past a defined point without an explicit decision is.

The distinction matters commercially, because both are described the same way in marketing. Asking whether the pause is structural or an instruction to a model separates the two, and the answers behave very differently on the day something goes wrong.

The cost of the design is latency: work waits. The benefit is that the class of mistakes that cannot be undone becomes impossible rather than unlikely. For anything binding, irreversible or expensive, that trade is usually correct.

See also approval gate, which is the specific mechanism, and agentic automation.