Human in the loop means a named person must review or approve certain AI decisions before they take effect. It works when the trigger points are defined in advance (by value, risk or uncertainty), the person sees the full history, and their decision is recorded. It fails when a person is asked to rubber-stamp everything.
George McKenna, Chairman and CTO, ETT Group. Updated 15 October 2026.
There are three common designs, and they suit different risks:
- Approve before action. The agent prepares, a person approves, then the action happens. Right for high-value or irreversible decisions.
- Exception routing. The agent handles routine cases alone and routes anything outside set limits to a person. Right for most back-office processes.
- Review after action. A person samples or reviews decisions afterwards. Right for low-risk, high-volume work, as long as the sampling is real.
What goes wrong is putting a person in the loop with no defined trigger, no context and too much volume. They end up approving everything, and the checkpoint becomes theatre. A good design keeps the routine volume away from people so the cases that reach them get proper attention.
Read more
the longer version
Show Your Working
A fortnightly piece on building and buying AI that pays its way. What I'm building, what it costs, and the questions to ask before you sign anything.