Type: url
Original: https://arxiv.org/pdf/2601.15195
Status: generated
Summary
A large-scale empirical study of 33,596 GitHub pull requests authored by five AI coding agents, quantifying which task types and PR characteristics predict merge success and deriving a hierarchical taxonomy (Reviewer/Pull-Request/Code/Agentic levels) from 600 manually-coded rejected PRs, finding reviewer abandonment as the single most common rejection cause.