AI coding · Workflow

Human-in-the-loop

Also called: HITL, Approval step, Confirm before run

The AI may propose or prepare actions, but critical steps wait for a human before they actually run—so it doesn't wreck production or delete data on its own.

In detail

Human-in-the-loop admits models make mistakes. Before dropping a database, changing payments, shipping to prod, or overwriting large business logic, hand the decision back to a person. "Ask before running a command" and "accept the Diff before write" in Agent tools are this idea.

It sits opposite fully automatic "YOLO" mode. Balance speed and risk: generating copy locally can be loose; credentials, permissions, and irreversible changes should be tight. Rules files can list what must pause for you.

For beginners: don't disable every confirmation to go faster. Seconds saved rarely buy back a hard-to-roll-back mistake.

Developer infoTerm ID, DOM cues, match priority

Term ID: ai-human-in-the-loop
DOM selectors: No DOM cues. This concept isn't detected directly on a page.
Priority: 1 · when several match at the same level, the higher priority wins
Version: v1 · updated Oct 5, 2026