AI Agent Adoption in Public GitHub Repositories
AI coding agents emerged in early 2025, and developers started adding configuration files to their repositories. This dashboard maps that adoption across 217,000+ public GitHub repositories scanned in Oct, 2025. The data shows actual usage patterns, adoption velocity, and tool preferences as they're unfolding in real projects.
Data boundaries to keep in mind:
- Public GitHub only – Private repositories and other platforms aren't visible here
- Configuration files are optional – Many teams use agents without config files, or choose not to commit them
- ⭐ Star range: 430K to 200 – GitHub's power law distribution makes scanning the lower star repos impractical, thus we scanned a subset. This range does capture the most visible and influential projects
Adoption Over Time
These are early days, the pattern is still forming. We are currently seeing Claude Code building momentum after its February launch, surpassing first-mover Cursor by July. Notice the Copilot uptick starting in August. Use the zoom and toggle features to explore adoption patterns across different tools and time periods.
Note: AGENTS.md (dashed line) represents an open standard for agent configuration, not a specific tool.
New vs Established Repositories
Forward-looking embrace vs. cautious retrofitting. Repositories created in 2025 show 13.7% adoption, three times higher than the 4.3% seen in established (pre-2025) projects. New projects face fewer legacy constraints and can build with agents from day one. I've started reaching out to select repositories to gather deeper insights on what's driving these adoption patterns and will be sharing findings.
New Repos (2025)
Established Repos (Pre-2025)
Single Tool vs. Multi-Tool: Still Experimenting
Most developers (86%) work with a single agent tool, while 13.4% are running multiple configurations side-by-side. The split suggests people are trying these tools in real projects, not just reading about them. Whether that 13.4% represents active experimentation or different team preferences is worth exploring further.