Dump-and-review culture is a development pattern in which AI generates a large code change and its author implicitly leaves verification to a reviewer. The apparent saving in implementation time moves work downstream to someone with less context and less time. section 1.4
The reviewer as implementer
A developer pastes a ticket into an agent, skips reading the diff, and feeds reviewer feedback back to the agent until the change merges. The reviewer effectively does the implementation work while the developer acts as an intermediary. section 1.4
The handoff diffuses responsibility: the author can assume the reviewer will check the code, while the reviewer must reconstruct its intent late in the process. Large generated diffs can encourage autopilot review and overreliance on green tests, leaving rare but critical invariant violations unnoticed. section 1.4 section 1.5
Unverified code creates trust debt: the deferred cost of understanding and checking it, often under pressure. Passing tests alone are not a safety case, especially when semantic questions—whether the change matches what the business meant—still require judgment. section 1.4 section 1.8
Replace the late handoff with verify-then-merge: write tests and contracts before generating code, then verify and take ownership of the result. A shared specification moves cognitive work earlier, where the team can reuse it. Adversarial AI review can surface contestable decisions, but overlapping model blind spots mean the loop must end with a human. section 1.8 section 1.5
