On July 5th, autonomous code agents (primarily Codex) authored around 15% of all public GitHub pull requests — and most of these changes were non-trivial. Wild times.

Does anyone know of any fully autonomous open-source agents that not only generate code on request, but also identify issues and open PRs in OSS projects without human prompting? That feels like a worthy experiment.

If benchmarks like SWE-Bench are accurate, these agents should already be able to resolve many issues in major Python packages — fully autonomously.

Data: https://insights.logicstar.ai/