Type: url
Original: https://arxiv.org/abs/2607.21832
Status: generated
Summary
This study characterizes agentic pull requests in comparison with human-generated pull requests and examines how their properties change across different stages of the software development lifecycle. Using the AIDev dataset, it analyzes how merge-rate differences between the two vary over time, identifies the development tasks AI coding agents are predominantly applied to, and compares key characteristics of the two kinds of pull request. It offers an empirical, longitudinal perspective on the role of AI coding agents in real-world software development.