For a growing share of the world's software, the first draft no longer comes from a person typing at a keyboard. GitHub's annual Octoverse report, published on October 28, says that by September 2026 developers started more than 40 percent of pull requests with a coding agent's first draft, up from 17 percent in March. That is 110 million agent-started pull requests in September alone, about six times March's volume.
A pull request is the basic unit of collaborative coding: a proposed change that teammates review before it is merged into a project. GitHub's figures cover the twelve months ending September 30, 2026, across public and private repositories unless noted. Developers pushed code 17.9 billion times in that year, more than in 2024 and 2025 combined, made 34.3 billion commits, and merged 1.3 billion pull requests, an average of 3.5 million a day.
GitHub is careful about what its headline number means. The report counts an agent signal as an observed sign of agent involvement in its telemetry, and calls that signal a floor, since not all agent use may be detected. The real share of agent-started work could therefore be higher than 40 percent. The figure also counts pull requests, not lines of code, and a merged change is not by itself proof that the change was correct.
Even with those caveats, the pattern inside the data is striking. Agent-started pull requests more often change several files at once, about 87 percent of the time, compared with 79 percent for pull requests without detected agent involvement. They are landing quickly, too. In February, 83 percent of agent-involved pull requests merged within two weeks, slightly behind the roughly 86 percent for the comparison group. By September, the agent figure had risen to 90 percent, while the comparison group held at about 86 percent.
The habit is spreading through the tools developers already use. Among GitHub Copilot's monthly active users, 55 percent now use agentic features, up from 35 percent a year earlier. The steepest growth is in the most hands-off mode: Copilot's cloud agent, where a developer assigns a task and the agent works on it independently, grew 230 percent year over year, against 85 percent growth for agent chat. Developers still ask agents questions, GitHub notes, but increasingly they hand over whole tasks.
Repositories built around agents from the start also look different. Among repositories created this year, the average one with an agent configuration file logged eight times as many commits as the average new repository without one. That is an association rather than proof that agents alone caused the difference, but it underlines how quickly agent-first workflows have become normal.
Model choice is fragmenting at the same time. In September, 61 percent of Copilot users who selected a model themselves used more than one model across their sessions, and developers on paid Copilot plans used 47 distinct models that month, up from 29 in May, with no single model carrying more than 25 percent of requests. GitHub now serves more than 225 million developers, led by the United States at 38.3 million and India at 32 million.
The bottleneck the report points to is no longer writing the first draft. It is review: judging what an agent produced, deciding whether it works, and deciding whether it belongs. GitHub closes on that note, observing that someone still has to come up with the idea and make those calls. What has changed, in one year of data, is how much of the drafting in between now starts with a machine.