Understanding the Rejection of Fixes Generated by Agentic Pull Requests -- Insights from the AIDev Dataset

The study analyzes the AIDev dataset to explore the rejection rates of code fixes generated by AI coding agents such as Copilot, Devin, Cursor, and Claude, revealing that 46.41% of proposed pull requests are rejected. It identifies 14 rejection reasons categorized into issues like incorrect implementations, failures in continuous integration tests, and low-priority fixes. This research highlights the necessity for improved guidance and prioritization in AI-generated code proposals to enhance efficiency and reduce resource wastage in software development processes.