By Hamza Faisal

The first draft of software is increasingly written by machines. GitHub's annual Octoverse report, published on October 28, says that by September 2026 more than 40% of pull requests on the platform started with a coding agent's first draft, up from 17% in March.

A pull request is a proposed change to a codebase that teammates review before it is merged, so the figure describes how work begins, not how much code agents write overall. In September alone, GitHub counted 110 million agent-started pull requests, about six times the volume it recorded in March.

Record activity across the platform

The surge sits inside a year of record activity. Developers pushed code to GitHub 17.9 billion times in the Octoverse year, which runs from October 1, 2025 to September 30, 2026, more than in 2024 and 2025 combined. Commits reached 34.3 billion and merged pull requests hit 1.3 billion, an average of 3.5 million a day.

GitHub also reports that new repositories with an agent configuration file logged eight times as many commits on average as new repositories without one. The pattern suggests agents are not just drafting individual changes but reshaping how whole projects get set up and run.

Review becomes the bottleneck

As drafting gets cheaper, reviewing becomes the scarce resource. GitHub says agent-started pull requests more often change multiple files, about 87% compared with 79% of pull requests without detected agent involvement, and 90% of agent-involved pull requests merged within two weeks in September, against about 86% for the comparison group.

That merge rate is a measure of acceptance, not quality. A merged change was approved by a human reviewer; it does not by itself prove the code was correct, secure or free of later rework. The practical shift is in the developer's job: less time typing the first version, more time specifying what is wanted, orchestrating agents and checking what comes back.

What the numbers do not prove

GitHub is careful about the limits of its own count. An agent-started pull request is one whose first draft came from a coding agent the company can detect, and GitHub describes the figure as a floor, since not all agent use may be detected. The true share of agent-started work is therefore probably higher, not lower.

The figures also describe GitHub's own platform, not the software industry as a whole, and they measure observed usage rather than model quality. Still, the direction is unmistakable: in six months, the agent went from starting fewer than one in five pull requests to starting more than two in five, and the bottleneck in software development has moved from writing code to reviewing it.