Understanding the Rejection of Fixes Generated by Agentic Pull Requests (AIDev dataset)

arXiv·high signal

Across the AIDev dataset, 46.41% of fixes proposed by Copilot, Devin, Cursor, and Claude are rejected. Analyzing 306 non-merged PRs, the authors identify 14 rejection reasons in four categories: incorrect implementation, CI pipeline failures, inability to execute the fix, and low priority. The work argues for better model guidance on implementation approach and self-validation before submitting.