Autonomy
How much an agent is trusted to act on its own before it has to stop and ask a human.
Think of it like
The difference between an intern who checks every email with you and a deputy you let sign things while you’re away.
Example
A coding agent with low autonomy shows each edit for approval; with high autonomy it edits, tests, and commits before you ever look.
How it actually works
Autonomy is a spectrum, not a switch — from suggest-only, to act-with-approval, to fully self-directed. More autonomy means more done unattended and more that can go wrong unattended. The practical craft is matching the level to the stakes: high for reversible, low-cost actions; a human gate for anything expensive or irreversible.
For product teams
The core trust dial — set it by how bad an unsupervised mistake would be.
For engineers
The degree to which the agent loop runs without human confirmation between actions.
Related
- Approval Gate — The checkpoint that lowers autonomy on risky steps.
- Agent Loop — More autonomy means longer unattended loops.
- Agentic Coding — A common high-autonomy application.
Read anything AI without the jargon
Look up any term in plain English, or save terms as you read with the free Chrome extension.
Open DecoderAdd to Chrome