GitHub reports 17.9 billion pushes in a year as developers adopt coding agents

GitHub, the Microsoft-owned code-hosting platform, published its annual Octoverse report, saying developers pushed code to GitHub 17.9 billion times in the 12 months to the end of September, more than in 2024 and 2025 combined. Commits across public and private repositories rose to 34.3 billion, which GitHub said is more than three times the level two years earlier. Merged pull requests more than tripled to 1.3 billion over the same period, an average of 3.5 million a day.

GitHub attributed the acceleration to agents moving from something developers were trying out to something they work alongside. Its figures showed activity rising steadily from January and bending sharply upward around June, which it linked to better underlying models, the harnesses built around them, and agents that run persistently and in the cloud. By September, GitHub said, commits were about six times their September 2025 level and merged pull requests about 3.9 times.

The agent-specific figures came from GitHub’s own detection. It reported that more than 40% of pull requests in September were started with an agent’s first draft, up from 17% in March, and that among repositories created in 2026 the average one with an agent configuration file logged eight times as many commits as one without. GitHub did not claim that the agent files caused the difference. It also said that 55% of paying Copilot users now use agentic features, up from 35% a year earlier, and that from June onward about 55% of all commits came from brand-new repositories.

The report carries its own caveats. GitHub said its agent detection is a floor, because not all agent use can be identified; that prior-year figures were recalculated on the same window and may differ from earlier reports; and that some industry classifications were produced by a language model. Pushes, commits and pull requests count activity, not value or quality, and the report does not address how much of the increase is substantive work. Its figures also come from the platform that sells a leading coding assistant, which readers may want to weigh.

For the record of AI-assisted software development, the report is one of the few platform-wide measurements of how much coding now begins with an agent, set against earlier evidence that was mostly from vendor surveys and individual studies.