Human Review
A person reading the work, with judgment as part of the job.
Review performed by a person: correctness in the real context, taste, maintainability, whether the thing should exist at all. The part of the process that automated checks cannot cover.
Review capacity becomes the throughput ceiling. The response is not to skip review but to make diffs smaller, intent clearer, and the mechanical parts verifiable by machine.
The scarce part of human review is attention, so spend it where machines cannot: design, naming, whether the abstraction is right, and whether the requirement itself is sound. Correctness of the mechanical parts is increasingly the automated check’s job, and duplicating it wastes the only resource that does not scale.
The review comment that mattered
An agent returns a working feature that is structured around a coincidence.
It passes tests and the tests are the problem — the design will not survive the next requirement. That is precisely what a person is for, and it is invisible to every automated check you could write.
- 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.
- reportState of AI-assisted software development 2025 ↗
Frames AI as an amplifier of the organisation it lands in: the same tooling helps strong delivery cultures and compounds problems in weak ones.
Human in the Loop · HITL
A person reviews or approves before the work lands.
Automated Review
A model reviewing a model’s output before a human sees it.
Automated Check
A machine verdict the agent cannot talk its way past.
Sycophancy
Agreeing with you because you sound sure.