Human in the Loop
also called HITL
A person reviews or approves before the work lands.
A workflow where a human stays in the decision path — reviewing a diff, approving an action, checking an output. The default posture for anything with real consequences.
The cost is throughput: attention becomes the bottleneck. The art is choosing which decisions genuinely need a person and automating the rest, rather than approving everything reflexively.
Place the human where judgement has leverage and the machine cannot verify: intent, scope, whether the thing should exist, and one-way doors. Placing them where a test could have decided is not safety, it is a bottleneck — and it trains people to approve without reading.
Reviewing the intent, not the diff
An agent completes a well-specified change and the tests pass.
The valuable human question is whether the change should exist and whether the spec was right — not a line-by-line re-derivation of what the tests already proved. That division keeps review meaningful instead of ceremonial.
- docsSafety best practices ↗
Application-level controls: adversarial testing, human review for high-stakes and code output, moderation, bounded inputs, and telling users what the model cannot do.
- docsUsing GitHub Copilot code review ↗
Details the boundary their own review agent draws by default: comments rather than approvals, with approval rights switched on deliberately.
- guidePeople + AI guidebook ↗
Human-centred design guidance for AI products: explainability, feedback, setting expectations and keeping a person in the loop. Cited as a reference — the page fetch returned only a landing stub, so do not quote specific claims here.
AFK · Away from keyboard
Running without a person watching.
Human Review
A person reading the work, with judgment as part of the job.
Permission Request
The agent pausing to ask before doing something risky.
Vibe Coding
Building by prompt-and-accept, without reading the code.