Developers pushed code to GitHub 17.9 billion times in the past 12 months, more than in 2024 and 2025 combined, according to the company's annual Octoverse report.
Across all repositories, public and private, commits (saved snapshots of code changes) reached 34.3 billion, more than three times the level of two years ago. Merged pull requests, which are proposed changes that a maintainer approves, more than tripled to 1.3 billion, or 3.5 million a day. GitHub Actions, the platform's automation service, ran 1,440 jobs every second, about 124 million a day. Public contributions hit 2.5 billion, against 1.12 billion last year.
The company ties the jump to agents, meaning AI programs that carry out coding tasks with limited supervision. Activity climbed from January and surged around June, it said, as agents moved from single sessions to persistent ones that track progress across days. As of September, developers start more than 40% of pull requests with an agent, up from 17% in March. In the same month, 61% of Copilot users used more than one model.
The headline comparison deserves care. New repositories with an agent configuration file logged eight times as many commits as those without one. That is a correlation. The teams that set up agent files early are likely the keenest adopters, and a tool that works in small, frequent steps will inflate commit counts on its own. Commits measure activity, not output.
Still, the shift in where human effort sits is real. The company describes the first move as directing an agent and review as the way developers shape the result. That makes review capacity, not typing speed, the likelier bottleneck.
The platform now serves more than 225 million developers. The United States (38.3 million) and India (32 million) each added more than 10 million this year, and Brazil added 3.28 million. One year into a five-year projection, the United States has closed 38.6% of the distance to its 2030 figure of 54.7 million, and India 28.3% of the way to 57.5 million. The company added that its own forecasting models no longer agree on what comes next, which is an unusual admission for a growth report.