Automated Review
A model reviewing a model’s output before a human sees it.
Using a separate agent or a fresh context to critique a change — reading the diff for bugs, checking against the spec, running extra tests — before it reaches a person.
Cheaper than human review and it catches a useful class of mistakes. It is not a substitute for independent verification, since a reviewer model can share the author’s blind spots.
Its value depends on independence: a fresh context, ideally a different model, and a reviewer that does not know what the author intended. A critic sharing the author’s context tends to inherit the author’s blind spots, which is why a second agent in a clean window finds more.
Review as a separate agent
An implementation agent proposes a change.
A second agent receives the diff, the spec and the tests, but not the implementation transcript, and is asked what could be wrong. It reliably catches the cases the author rationalised away, because it never saw the reasoning.
- 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.
- blogBuilding effective agents ↗
Draws the line between fixed workflows and agents that choose their own process, and names the core patterns: chaining, routing, parallelisation, orchestrator-workers, evaluator-optimizer.
Automated Check
A machine verdict the agent cannot talk its way past.
Human Review
A person reading the work, with judgment as part of the job.
Agent Mode
The mode where the assistant acts instead of suggesting.
LLM as Judge · Model grading
Using a model to score another model’s output.