Type: url
Original: https://arxiv.org/pdf/2604.03196
Status: generated
License: CC-BY-4.0
Summary
An MSR ‘26 empirical study analysing 3,109 pull requests drawn from the AIDev dataset to compare merge outcomes between human-only reviews and reviews performed solely by code review agents (CRAs). It reports that CRA-only PRs merge at 45.20% against 68.37% for human-only PRs, with a 34.88% abandonment rate, and that 60.2% of closed CRA-only PRs fall into the 0-30% signal-to-noise band while 12 of 13 CRAs average below 60% signal. It concludes that CRAs should augment rather than replace human reviewers.