What is happening with code reviews?

Engineering organizations are changing code review because AI agents can create more and larger pull requests than people can inspect line by line. Gergely Orosz reports several emerging workflows: people review the findings from AI reviewers, teams classify changes by blast radius, and reviewers focus on plans, tests, database schemas, or the before-and-after behavior rather than every implementation detail. OpenAI and Anthropic told Orosz that they use risk-based review, with higher-risk changes receiving human attention. Some smaller teams allow low-risk changes through automated checks, but Orosz found more discussion of eliminating human review than evidence that it is common. The practical shift is from uniform inspection to selective assurance. That can shorten queues, but it moves more responsibility into risk classification, tests, rollout controls, and observability. Teams adopting the pattern should define which surfaces always require a person, then verify that their automated safeguards catch the failures a line-by-line review used to catch.