There is more to code review than (automatable) detection
An agent can process a diff. It can’t say ‘we had an incident here last Tuesday.’ John Allspaw on code review as coordination and governance, not just defect-finding.
A recent paper argues that every stated goal of code review can be served by agents at lower cost and higher throughput. John Allspaw's answer is that the paper broke review into detection tasks and left out what reviewers bring: we just had an incident in this service, that interface is about to be deprecated, legal told us to stop logging that field. Review is also coordination, sensemaking and governance, and an agent that signs off on a pull request bears no consequences for it. If you are being asked to replace reviewers with agents, list which of those jobs your review actually does before deciding which ones you can drop.
Comments
Loading comments…