Detecting AI Coding Agents in Open Source: A Validated Multi-Method Census of 180 Million Repositories

A multi-layered detection framework has been developed to analyze AI coding agents across 180 million Git repositories, revealing significant insights into their prevalence and activity. The study identified 850,157 commits attributed to Claude Code, with a notable detection gap where traditional methods underestimated agent presence by a factor of 30. This research highlights the limitations of single-method detection approaches and underscores the importance of multi-method strategies for accurately understanding AI agent contributions in open-source projects, as different detection channels capture distinct populations and types of work.