07 / Ways of Working
Human-in-the-loop
Working alongside the agent while it runs, reviewing, correcting and steering. The slowest pattern per task, and by far the fastest way to learn what the thing is actually good at.
How you’ll hear it
“Should we automate the approvals?”
“Not yet. Keep a person in the loop until we trust it.”
Avoid
Calling it “human in the loop” when the human only clicks approve.
The longer version
Human in the loop means a person actively reviews, corrects and steers while the work happens. It is slower per task and the fastest way to learn where AI can be trusted. Meaningful oversight needs time and information, not just a button.