Martin Monperrus

On 11 June 2026, Martin Monperrus, a software engineering researcher at KTH, posted a position paper to arXiv arguing that human code review is no longer a necessary component of a quality pipeline. The title states the claim without hedging: "The End of Code Review: Coding Agents Supersede Human Inspection".

Two arguments carry it. LLM-based autonomous systems meet the traditional goals of review (finding defects, improving quality, ensuring conformance, sharing knowledge) at lower cost and higher throughput, without human fatigue or inconsistency. And the dominant hybrid arrangement, where the agent writes and the human reads, provides neither real quality assurance nor the ability to keep pace with AI-assisted production. What it does produce, Monperrus says, is false security. He aims at Fagan inspection (1976) and leans on Bacchelli and Bird, whose work shows review catches fewer bugs in practice than developers believe; SWE-bench resolution rates of roughly 20 to 40 percent serve as the capability evidence.

His replacement is an adversarial multi-agent verification pipeline: a generating agent, independent reviewer agents, a verification layer adding automated tests and formal methods, and a consensus vote to accept or reject. Humans move to specification, architecture arbitration, critical domains, edge cases, and final approval on sensitive systems.

The recommendations sit oddly against the thesis: pilot agent review on low-risk components first, keep an initial hybrid workflow where agents flag and humans approve. SFEIR's Review episode of 30 July 2026 takes the diagnosis and refuses the conclusion: acceptance does not get delegated.