Semi-Autonomous
An agent that does the work on its own but pauses to ask a human before the risky moves.
Think of it like
A learner driver — hands on the wheel most of the time, but the instructor grabs it before anything expensive happens.
Example
The coding agent writes and tests the fix by itself, then stops and asks “push to main?” before it actually deploys.
How it actually works
The design question is where to put the gates — too many and it’s just a slow assistant, too few and it’s effectively autonomous. Good semi-autonomous systems reserve the human for irreversible or high-cost actions and let the agent own everything cheap and undoable. It’s the pragmatic middle most production agents actually live in.
For product teams
You get most of the speed of autonomy while keeping veto power over the moves that can hurt.
For engineers
The loop inserts a blocking approval step before actions flagged as high-risk or irreversible.
Related
- Approval Gate — The checkpoint where it waits for a yes.
- HITL — Keeping a person on the consequential calls.
- Autonomous Agent — Sits between manual and fully unsupervised.
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