What's this about?
An empirical study of how code review changes as generative AI joins the process — comparing review quality across successive generations of AI technology as review shifts from a human process to an AI-supported and increasingly agent-driven one (HF 11↑).
What exactly is "code review"?
Before code joins a shared codebase, another developer reads and critiques it — catching bugs, style problems, and design mistakes. It's one of software's main quality gates and one of its biggest time sinks, which is why it's a prime target for AI: agents can now review pull requests automatically.
Why does it matter?
Review is where AI-written code meets human oversight; if AI reviews AI-written code, the oversight question compounds. Measuring whether newer AI generations actually review better (rather than just faster) is exactly the evidence that debate needs.