Human in the Loop
Manusia dalam Lingkung
A person who reviews, approves, or corrects the agent's work at key points.
Human in the loop means a person stays in the path of the work: approving a risky action, reviewing the output, or correcting the agent when it drifts. The amount of human involvement is a design decision, not a moral one.
There are four useful positions. The human approves every action (safe, slow). The human approves only irreversible or outward-facing actions (the usual choice). The human reviews the result at the end (fast, catches errors late). The human is available on request (fastest, weakest guarantee).
The failure mode of the middle position is approval fatigue: if you are asked to approve twenty safe actions and one dangerous one, you will click through all of them. The fix is to make the prompt itself carry the risk information, so the reviewer's attention goes to the action that deserves it.
Related terms
Agent
A model with tools, a system prompt, and a loop that lets it act and observe results.
Guardrails
Checks that stop or constrain an action before it happens.
Approval
The moment a person says yes before the agent does something it cannot undo.
Agent Mode
A preset that bundles permissions and behaviour, like a profile for the agent.
Code Review
A second pass by someone who did not write the code, looking for what the author missed.
Want this working inside your team?
It starts with one conversation about your workflow.