The signal-to-noise ratio on public GitHub is shifting under our feet. According to a new report from MyClawn, AI coding agents opened 427,042 pull requests on public repositories between November 11 and November 12, 2026. That single dayโ€™s output represents 31.9% of the total 1,339,893 pull requests opened on the platform during that window. This isn't a survey or a vendor press release; it is a raw count derived from GitHub's public search API, filtered by the specific branch prefixes and description markers that the four most widely used agents leave on their work.

The Illusion of High-Quality Automation

On the surface, the numbers look like a triumph for autonomous coding. Over a 30-day period, agents submitted 10,862,952 pull requests, of which 91.0% were merged. However, digging into the metadata reveals a different reality. Of the latest 115 agent pull requests analyzed, 63% were opened by the repository's own owner. Another 13% came from collaborators. Only 24% originated from outside the core team. This suggests that the high merge rate is largely an artifact of developers using agents to push changes to their own repositories, where they merge their own work minutes after generation, bypassing traditional peer review.

The Open Source Stress Test

The dynamic changes drastically when looking at large open source projects that accept contributions from strangers. The repository openclaw/openclaw received 806 agent pull requests in 30 days, accounting for 4% of its total 22,310 incoming PRs. Of those, 492 were merged and 125 were closed without a merge. Notably, 29 of the last 30 agent PRs came from outside the team, indicating that while agents are flooding open source with proposals, maintainers are still exercising significant judgment, rejecting a substantial portion of external agent contributions. The merge rate for external agents is far more selective than the self-merge rate implies.

Methodology and Limitations

The count is a floor, not a census. Private repositories are invisible to this metric, as is any agent work that lacks the specific branch prefixes or description markers used for identification. Furthermore, the method double-counts pull requests that carry marks from multiple agents in the 30-day aggregate, though it deduplicates them for the 24-hour snapshot. A human could theoretically mimic these markers, and bots can open PRs under accounts that cannot be tied to a specific person. Despite these limitations, the data provides a rare, unfiltered look at the sheer volume of automated code submission hitting public infrastructure.

Key Takeaways

  • AI agents opened 427,042 public GitHub pull requests in 24 hours, representing 31.9% of all daily PR activity.
  • The 91% merge rate is skewed by self-merges; 63% of recent agent PRs were opened by repository owners.
  • External agent contributions to open source projects like openclaw/openclaw face higher scrutiny, with significant rejection rates.
  • The data is derived from GitHub's public search API using specific branch prefixes and description markers left by four major coding agents.

The Bottom Line

The 91% merge rate is a vanity metric driven by self-merges in private workflows, not a sign of autonomous code quality. Real-world open source integration remains a bottleneck where human maintainers still hold the gate.