← Concept Index

Agents — core

Autonomy levels & approval gates

Also called: human-in-the-loop, checkpoints

DEFINITION

Autonomy is how far an agent can go before a human must confirm. An approval gate pauses the loop before a consequential action — sending, paying, deleting — and asks a person to approve it.

WHY IT MATTERS

The gate is the single most important safeguard: it's what stops an injected or mistaken instruction from becoming an irreversible action. Deciding which actions need a gate is a design choice, not a default.

COMMONLY CONFUSED WITH

Trusting the model more. Higher autonomy isn't a reward for a good model; it's a risk decision about what a mistake would cost.