Term: Human-in-the-loop
~5 min read
Estimated time: ~5 min read — for the in-app brief plus opening the primary source.
What this is
Human-in-the-loop means a person reviews or approves before the AI’s output becomes an action. Specify when, who, and on what.
Everyday example
A refund bot drafts the decision; a supervisor must approve before money moves. That approval is human-in-the-loop — not a slogan, a step.
Human-in-the-loop means a person must review or approve before the AI’s output becomes an action.
- It is a control, not a slogan — specify when, who, and on what.
- Spot-checks on 2% of tickets is not the same as approval on every refund.
- Over time, people rubber-stamp. Design against that.
- Regulated people-decisions almost always need a human decision-maker.
Next action: Write the human step for your highest-risk AI workflow: who, before what action, with what log.
What changes in how you lead
How decision rights, process, and ownership should change.
- HITL is in the process design, not in a footnote.
- HR and Legal decide which decisions may never be fully automated.
Compare related ideas
Human-in-the-loop vs Agent
An agent acts. Human-in-the-loop is the control that stops some of those actions until a person says yes.
Open AgentDeep dive
Spot-checking 2% of tickets is not the same as approving every payment.
People drift into rubber-stamping — design against that (samples, time, accountability).
Regulated people-decisions almost always need a human decision-maker.
Write the step into the process, not into a footnote.