More than 40% of the pull requests opened on GitHub in September 2026 started with a first draft written by an agent. In March it was 17%. That single number sums up the Octoverse 2026 report, presented on October 28, 2026 at GitHub Universe, and it explains why almost every other metric tripled in a year.
The numbers
The report covers the twelve months to September 2026. The volume itself is not the story, since it always grows; the speed is. Much of the jump landed in the second half of the year, right as coding agents moved from experiment to everyday tool.
| Metric | Octoverse 2026 | Change |
|---|---|---|
| PRs started by an agent | Over 40% (September) | 17% in March 2026 |
| Agent-started PRs in one month | About 110 million (September) | Roughly six times March |
| Merged PRs (12 months) | 1.3 billion, 3.5 million a day | More than tripled |
| Merged PRs per month | 220 million | Tripled in under a year |
| Commits in September | About 7.38 billion | Over three times a year earlier |
| Commits and pushes (12 months) | 34.3 billion commits, 17.9 billion pushes | — |
| New repositories per month | 30.3 million | Doubled |
What agent-started PRs look like
The report offers two clues. First, 87% of agent-started PRs touch more than one file, against 79% of the rest. They are broader changes, and broader changes need more context to review.
Second, repository setup matters. Among repos created in 2026, those with an agent configuration file averaged eight times as many commits as those without one. That is correlation, not causation: a team that bothers to write the file is probably already leaning hard on agents. Still, it shows those files are no longer a curiosity.
TypeScript wins with agents
GitHub also looked at which languages agent users pick. Among repos created in 2026 that show agent activity, TypeScript is the primary language in 35%; without that signal, it drops to 24%. The authors' explanation matches day-to-day experience: with static types, the compiler catches mistakes in generated code early, so checking what an agent wrote is cheaper.
What it means
For maintainers the takeaway is blunt: writing code is no longer the bottleneck, reviewing it is. If PR volume triples and each PR touches more files, human review cannot keep pace. What does scale is automation: tests that actually cover the change, linters and type checks in CI, branch protection rules and well-defined code owners.
There is an infrastructure bill too. Three times the commits and merges means roughly three times the CI runs, builds and deployments. Teams paying for runner minutes or living with a slow pipeline will feel it first.
And for anyone choosing a stack, the TypeScript figure points to an advantage benchmarks don't show: types and a strict compiler make agent output easier to validate.
What we still don't know
The figures count PRs that an agent starts, not how many get merged unchanged or reverted later. Without that rate there is no way to tell whether quality is keeping up with volume, and that is the question that matters to a team.
Sources
- Octoverse: Commits, pushes, reach new highs in 2026 as developers build with agents (GitHub Blog, October 28, 2026)
- GitHub Octoverse 2026 (Simon Willison, October 28, 2026)