🇺🇸 English 🇰🇷 한국어 🇯🇵 日本語
Home › Developer Tools
Developer Tools

Agents Now Start 40% of GitHub Pull Requests

AI agents started more than 40% of GitHub pull requests in September 2026, up from 17% in March, GitHub's Octoverse report said Oct. 28, as the company launched Copilot Fusion at its Universe conference.

AI agents started more than 40% of pull requests on GitHub in September 2026, up from 17% in March, according to GitHub's annual Octoverse report released Oct. 28. The figure landed the same day GitHub opened its Universe conference in San Francisco with Copilot Fusion, a system that picks which AI models handle a coding task.

Lines of CSS code displayed on a dark computer screen
CSS code on a computer screen, a file photo. 사진: Sai Kiran Anagani _imkiran / 위키미디어 공용(CC0 1.0)

Key takeaways

  • Agents started about 110 million pull requests in September, roughly six times March's volume, GitHub said.
  • Merged pull requests reached 1.3 billion in the Octoverse year to Sept. 30, up from 538 million a year earlier.
  • Copilot Fusion routes work across models; GitHub claims its Max level nears Claude Opus 5.5 Max at half the cost.

How many GitHub pull requests do AI agents create?

GitHub's Octoverse year runs from Oct. 1, 2025, to Sept. 30, 2026. Within it, the share of pull requests that began with an agent rose from 17% in March to more than 40% in September. Agents opened about 110 million pull requests in September alone.

Commits nearly tripled and GitHub Actions jobs more than tripled, as the table shows.

GitHub activity by Octoverse year (12 months to Sept. 30), per GitHub
Metric202420252026
Pushes5.2B6.7B17.9B
Commits10.0B12.9B34.3B
Merged pull requests402M538M1.3B
GitHub Actions jobs8.0B14.0B45.4B

Copilot users are also leaning on agents. GitHub said 55% of Copilot monthly active users now use agentic features, up from 35% a year earlier, and Copilot cloud agent usage grew 230% year over year.

What is GitHub Copilot Fusion?

Copilot Fusion decides how to run a task across one or more models. Developers choose an effort level, Standard, High or Max, according to The New Stack. Fusion can hand a job to one model, escalate a weak answer to a stronger one, or have one model critique another's work.

In a keynote post, Jay Parikh, Microsoft's executive vice president of CoreAI, said Fusion Max scored within two points of Claude Opus 5.5 Max on the Artificial Analysis Coding Agent Index. GitHub says it did so at half the cost and with 40% less time per task. Those are GitHub's own figures. Administrators can limit which models Fusion may choose.

In September, 61% of Copilot users who picked a model used more than one, GitHub said.

Copilot code review Max and enterprise pricing

GitHub added a Max tier to Copilot code review, alongside the existing Lite and Balanced tiers. It said Max posted the highest F1 score on ReviewBench for high-severity issues.

Starting Oct. 28, GitHub Enterprise customers can use Copilot features, including the app, code review and automations, without a separate Copilot subscription. They pay for the AI credits they use. The plan is sold through GitHub Sales now, with self-serve "coming soon."

Context: building for agent-scale traffic

GitHub said it has added 4 million CPU cores, 160 petabytes of storage and 50 terabits per second of network capacity so far in 2026. It expects to finish a move to a multi-region Azure deployment early next year.

"Completed work, not more code, is the output that matters," Parikh wrote. Rivals are targeting the same shift; former GitHub CEO Thomas Dohmke's startup Entire this week opened a code forge built for agents. Faster output also raises review stakes, as earlier research on LLM bug-fixers that broke working code showed.

What's next

Universe runs through Oct. 29. The test for Fusion will be whether independent benchmarks confirm GitHub's cost claims, and whether review tools keep pace as agent-written pull requests approach half of all activity.

FAQ

What share of GitHub pull requests come from AI agents?

More than 40% of pull requests in September 2026 started with an agent, up from 17% in March, according to GitHub's Octoverse report.

Is Copilot Fusion cheaper than Claude Code?

GitHub says Fusion Max costs about half as much per task as Claude Code with Opus 5.5 Max while scoring within two points. No independent test has confirmed this yet.

Do GitHub Enterprise customers need a Copilot subscription?

Not from Oct. 28. Enterprise customers can use Copilot features and pay for the AI credits they consume.

References

#GitHub #Octoverse 2026 #GitHub Universe #Copilot Fusion #AI coding agents #pull requests #Copilot code review #Jay Parikh #GitHub Enterprise #AI credits
S
Seung Jung
AI Newsway · Developer

Editor responsible for AI Newsway. Writes about AI, developer tools, cloud and productivity technology.

More by this reporter ›
Copyright © AI Newsway All rights reserved. Unauthorized reproduction, redistribution, and AI training prohibited.
Share
f X K

Comments 0

Comments are currently disabled.

Be the first to comment.