Human-in-the-Loop — Half Dev Glossary

In short

A design in which a person reviews, approves or corrects an AI system's work at defined points instead of letting it run fully on its own.

Human-in-the-loop means a person is part of the process, not only a reader of the final result. The AI may draft, classify or propose an action, and a human approves, edits or rejects it before it takes effect. The review point can sit before an irreversible step, such as sending a payment or publishing a page, or it can sample a share of outputs for quality checks.

A practical example is an AI that drafts replies to customer emails. Routine, low-risk replies might go out after a quick glance, while refunds, legal questions or upset customers are held for a person to edit. Deciding which cases need review, and showing the reviewer the evidence behind the suggestion, matters as much as having a reviewer at all.

Human review has costs. It adds delay, and it can become a rubber stamp when people see too many items. Track how often reviewers change the AI's output, and turn those corrections into test cases for AI agent evaluation.