What it means for leaders

Coding agents now open pull requests on their own at scale, not just suggest lines of code. Decide how agent work enters your repositories: who assigns it, who reviews it and what it may merge.

What would change the reading: Month-over-month growth by agent, and shifts in which agent leads.

The trend

June 30, 2025September 30, 2026

The breakdown

September 30, 2026.
  • OpenAI Codex938,679
  • GitHub Copilot60,007
  • Devin39,893
  • Cursor31,851
  • Claude19,487
  • Google Jules10,148

Every reading

DateValueChangeSource
1,100,065+20.5%Source
912,926+13.6%Source
803,968+23.8%Source
649,210−17.1%Source
782,933+6.0%Source
738,422−10.0%Source
820,637+20.9%Source
678,574+25.6%Source
540,358+13.4%Source
476,309+0.1%Source
475,865−15.2%Source
561,320+24.0%Source
452,746+3.4%Source
437,651+45.1%Source
301,681−24.7%Source
400,812—Source

Definition and method

What it measures
Pull requests opened in the month by cloud coding agents, counted by GitHub search for each agent's account or branch pattern. Parts give each agent.
Population
Pull requests opened in public GitHub repositories
Updated
Monthly

Caveats

  • Public repositories only.
  • Agents are identified by the app account that opens the pull request, or for OpenAI Codex by its "codex/" branches, which some people also use by hand; agent work submitted through a person's own account (most local coding assistants) is not counted.

Cite this

Enterprise AI in Numbers, "Pull requests opened by AI coding agents on public GitHub, per month", https://lucianoayres.com/ai-brief/data/agent-authored-prs/, retrieved October 25, 2026. Data: GitHub search API, our monthly counts.

Download every indicator (CSV) · Methodology