Agents
GitHub Octoverse 2026: AI agents now start more than 40% of pull requests
GitHub's Octoverse 2026 says agents drafted over 40% of pull requests in September, up from 17% in March. What the numbers measure, and what they don't.
In March 2026, 17% of pull requests on GitHub started with a draft written by a coding agent. By September, it was more than 40%. That is the headline figure of GitHub's annual Octoverse report, published on October 28, 2026 alongside the GitHub Universe announcements (GitHub).
These are GitHub's own platform measurements. GitHub also sells Copilot and its coding agents, so the report is useful data on how its platform is used, not an independent audit.
The key numbers
- Pull requests started by an agent: more than 40% in September 2026, against 17% in March, or about 110 million in September alone, roughly six times March's volume.
- Overall activity: 17.9 billion pushes over the 12 months to September 30, 2026, more than in 2024 and 2025 combined, 34.3 billion commits and 1.3 billion merged pull requests (about 3.5 million a day).
- Agent use among Copilot users: 55% of Copilot's monthly active users use agentic features, against 35% a year earlier.
- Several models: in September, 61% of Copilot users who choose their model themselves used more than one, up from 50% four months earlier.
Source for all figures: GitHub's Octoverse 2026 post.
What "started by an agent" means
GitHub counts a pull request as agent-started when its first draft came from a coding agent that GitHub can detect. A developer may then rewrite part of it: it still counts. GitHub calls the figure a floor, because some agent use goes undetected.
So the 40% does not mean agents wrote 40% of the code merged on GitHub. The report does not say how much of each agent draft survives review, nor break the figure down by public and private repositories.
Are agent pull requests merged as often?
According to GitHub, about 87% of agent-started pull requests change several files, against 79% of the others. In February, 83% of agent-involved pull requests were merged within two weeks, slightly behind the roughly 86% for the rest. In September, the agent figure reached 90%, while the comparison group stayed around 86%.
A merge rate measures whether a change was accepted, not whether it was correct or needed rework later. GitHub also notes that new repositories with an agent configuration file logged eight times as many commits as those without one. That is a correlation, not proof that the file causes the extra activity.
Why review becomes the bottleneck
If agents draft a growing share of changes, the human work moves to specifying the task and reviewing the result. GitHub's own launches the same week point that way: Copilot Fusion picks and combines models for each task, and Agent Merge watches pull requests for failed checks and merge conflicts.
The report also says usage of AI-assisted code review and GitHub Advanced Security each grew more than 90% year over year, and that push protection, which blocks exposed secrets, ran 5.8 billion scans this year.
Languages and countries
TypeScript remains the most used language on GitHub by contributors, ahead of Python and JavaScript. GitHub now counts more than 225 million developers. The United States (38.3 million) and India (32 million) are the two largest communities; France has 4.75 million.
What to keep in mind
- The 40% figure is a share of pull requests whose first draft came from a detectable agent, measured by GitHub on its own platform.
- It is a floor, not a measure of code quality or of how much agent code ends up in production.
- For teams, the practical question is less "should we use agents" than how to review their output: tests, required checks and clear task descriptions.
Spotted an error in this article? Report it · Our corrections policy
