First longitudinal empirical study of the post-merge lifecycle of agentic and human code in real-world open source projects.
182public GitHub repositories
30K+agentic commits
121M+lifecycle lines tracked
12maintenance operational intents
143K+security & dependency findings
1 yearMay 2025 to May 2026
Finding Highlights:
46%
higher corrective maintenance rates than human contributions
1.51x
times more high-severity static security findings than human contributions
Agentic code share
increases the corrective maintenance burden within projects
No review rate
increases the agentic maintenance burden of a project
Interested to learn more? Check out the interactive visualization to play with the data yourself!
Citation:
@article{postmerge26,
title = {Do These Violent Delights Have Violent Ends?
Measuring the Post-Merge Fate of Agentic Code},
author = {Xia, Chunqiu Steven and Miller, Courtney},
year = {2026},
journal = {arXiv preprint arXiv:temp},
url = {https://arxiv.org/pdf/temp}
}
Repo-Week Timeline
Weekly tracked-line survival rates for agentic and human origin lines.
Commit Contribution Maintenance Class Mix
Origin-to-Terminal Actor Flow
Line-Level Termination Maintenance Class Proportion