GitHub Octoverse 2026: Agents Now Start Over 40% of Pull Requests, Up From 17% in March

TL;DR

GitHub's Octoverse 2026 report, published October 28 during GitHub Universe, puts a number on the agent takeover of the pull request. By September 2026, more than 40% of PRs on GitHub started as a coding agent's first draft, up from 17% in March. That is 110 million agent-started PRs in one month, about six times March's volume, and GitHub calls the figure a floor because it only counts agents it can detect. The volume underneath moved with it: developers pushed code 17.9 billion times in the Octoverse year, more than 2024 and 2025 combined, commits reached 34.3 billion, merged PRs hit 1.3 billion, and GitHub Actions ran 45.4 billion jobs. GitHub now counts more than 225 million developers. TypeScript held first place among languages, Rust doubled, and developers on paid GitHub Copilot plans used 47 different models in September with no single model above 25% of requests.

The pull request now starts with an agent, by default

The headline sentence, verbatim: "By September 2026, more than 40% of pull requests on GitHub started with an agent's first draft, up from 17% in March." GitHub adds: "That is 110 million in September alone, about 6x March's volume." The glossary defines an agent-started PR as "a pull request whose first draft came from a coding agent that we can detect," and the methodology section is blunt about the limit: "It is a floor, since not all agent use may be detected."

An agent that commits through a developer's own account and leaves no trace in telemetry is a ghostwriter with the author's name on the cover, and GitHub can only count the ghostwriters who sign their own name. So 40% is the detectable share. The real share is higher, and nobody outside GitHub can say by how much.

The agent PRs also look different. About 87% of agent-started pull requests change multiple files, against 79% for PRs with no detected agent involvement. Their merge rate moved as well: in February, 83% of agent-involved PRs merged within two weeks; by September that was 90%, while the comparison group held at about 86%.

Three years of commits in one

Every top-line counter GitHub publishes roughly tripled against its own 2025 number, with the 2024 and 2025 figures recalculated on the same definitions and window.

• Pushes: 5.2 billion in 2024, 6.7 billion in 2025, 17.9 billion in 2026. GitHub: "Developers pushed code changes to GitHub 17.9 billion times in the past 12 months, more than 2024 and 2025 combined."
• Commits across public and private repos: 10.0 billion, then 12.9 billion, then 34.3 billion.
• GitHub Actions jobs: 8.0 billion, 14.0 billion, 45.4 billion.
• Merged pull requests: 402 million, 538 million, 1.3 billion, about 3.5 million a day. Merged Dependabot PRs alone passed 30 million, roughly double the 2025 pace.

The curve is not smooth. Month over month, from September 2025 to September 2026, commits grew about 6x, Actions jobs about 4.8x, pushes about 4.5x and merged PRs about 3.9x, and most of that came in the back half. GitHub: "Our telemetry shows activity climbing steadily from January, then surging around June." It attributes the surge to how the tools changed: "Agents moved from single-session tools to persistent ones that could track progress across days," and "agents began making extensive use of the cloud, executing work remotely as well as on local machines."

From June onwards, about 55% of all commits came from brand-new repositories. And among repos created this year, "the average repo with an agent configuration file logged eight times as many commits as the average new repo without one." That is a correlation in a self-selected population, and GitHub does not claim otherwise, but if you have been putting off writing the AGENTS.md, there is your excuse.

Simon Willison posted the keynote charts from Universe: 7.38 billion commits in a single month, 220 million merged PRs in a month, and 30.3 million new repositories a month, with the new-repo rate doubling over the year.

Nobody is loyal to one model

Among Copilot's monthly active users, 55% now use agentic features, up from 35% a year ago. Copilot cloud agent usage grew 230% year over year and agent chat grew 85%. The more useful number for anyone deciding what to build on is what happens when developers choose the model: in September, 61% of users who picked a model used more than one, up from 50% four months earlier. Developers on paid plans used 29 distinct models in May and 47 in September, "and no single model carried more than 25% of requests that month."

Out of every 100 developer-selected requests, 38 went to general purpose models, 37 to deep reasoning and planning models, and 25 to fast and lightweight models. The models named in the report include Claude Opus 5.5, GPT 5.5, Claude Sonnet 5, Grok 4.7, GPT 6.1 Sol, Claude Haiku 5.5, Gemini 3.8 Flash and GPT 6 Luna. 63% of paying developers used auto model selection at least once in September, and GitHub's HydraFusion router, which routed almost 60,000 sessions in the CLI and the Copilot app, is now generally available as GitHub Copilot Fusion.

The model figures count only developer-selected requests. They exclude auto selection, staff usage and anyone bringing their own API key, and GitHub says they "show observed use, not model quality."

TypeScript stays on top, Rust doubles, Java sits out

The language order did not move: TypeScript first, Python second, JavaScript third. Monthly contributors grew by 3.62 million for TypeScript, 3.23 million for Python and 1.82 million for JavaScript, the largest gains in the top ten. Rust "grew fastest by percentage among the top languages, doubling in the past year." GitHub's forecast for September 2027 has TypeScript at about 13.07 million monthly contributors and Python at about 12.56 million.

The new cut this year is how much of each language's merged PR volume carries a detected AI signal. TypeScript leads at 44.1%, Python is at 41.1%, Rust at 40.4%, and Java is lowest at 21.2%. Among repositories created in 2026, TypeScript was the primary language in 35% of those with an agent signal versus 24% of those without.

225 million developers, and India closing on the US

GitHub counts more than 225 million developers. The United States has 38.3 million and India 32.0 million as of September 2026, and each added more than 10 million this year. China is third at 12.4 million and Brazil fourth at 10.2 million after adding 3.28 million. Digit worked the growth rates out of the report: India grew about 46% against about 37% for the US, though the absolute gap widened slightly, from 6.1 million to 6.3 million. GitHub's 2030 projections put India at 57.5 million developers and the US at 54.7 million.

GitHub ran 43 forecast models for growth through September 2031, up from five last year, and 21 of them project the developer population tripling or more. Public repositories grew from 9.0 million in 2022 to 25 million in 2026, and GitHub's industry classifier, which sorts public repos into NAICS sectors with a language model, found that construction has the highest share of agent-opened PRs of any industry, at about 9%. GitHub flags the industry shares as estimates.

What developers starred this year

GitHub split the fastest-growing repos into two lists. Community and independent projects: cc-switch, awesome-mcp-servers, omarchy, first-contributions and llama.cpp. Single-vendor and product-led repos: claude-code, codex, opencode, vscode and copilot-cli. Four of those five are coding agents and the fifth is the editor they run inside.

A few smaller counters point the same way. Public repositories mentioning interoperability or reverse engineering, found by terms like "clean room implementation" and "drop-in replacement," rose from about 30,000 in 2025 to over 130,000 in 2026. AI-assisted code review and GitHub Advanced Security usage each grew more than 90% year over year, and push protection ran 5.8 billion secret scans, nearly four times the 2025 pace. Stacked pull requests, which split a change into smaller sequential PRs, launched as a preview on July 30, and more than 264,000 developers had used them by September. The most-used emoji across every industry was the rocket, which is either optimism or the deploy pipeline asking for help.

Read the fine print

This is GitHub measuring GitHub, and the report is more careful about that than the headlines will be. The agent share is a floor. "AI involvement" is built from two signals, AI mentions and agent-opened pull requests from detectable agents only. The Octoverse year now runs October 1 to September 30 and includes private repositories, where previous reports covered public activity from September 1 to August 31, so you cannot line these figures up against last year's published report without GitHub's recalculated baselines. A push is an upload event, not a quantity of code: "A push sends one or more commits to GitHub," and one agent committing after every edit inflates both counters without writing more software.

The section GitHub titled "What doesn't change" is short and reads as if someone in the building wanted it on the record: "None of that activity changes who decides what the code should do." And: "someone still has to decide whether it works, whether it belongs, and whether it's good." As far as anyone can detect, an agent did not write that paragraph.

What this means if you ship code

• Review is the constraint. 110 million agent PRs a month, 87% of them multi-file, and 90% merging within two weeks means the humans are either reading faster or reading less. Stacked PRs and AI-assisted review both grew this year; decide which one your team is doing on purpose.
• Write the agent config file. Repos with one logged eight times as many commits as repos without. Correlation, yes, but it is also the cheapest thing in this report to copy.
• Do not hard-wire a model. 47 models in a month, none above 25%, and 61% of model-pickers using more than one. Make the vendor a setting.
• TypeScript, Python and Rust are where the agents live, at 40% to 44% AI involvement in merged PRs. Java at 21% is a different world for now.

Key Takeaways

• GitHub's Octoverse 2026 says more than 40% of pull requests started as a detectable agent's first draft by September 2026, up from 17% in March: 110 million agent-started PRs in one month, about 6x March.
• Pushes hit 17.9 billion (5.2 billion in 2024, 6.7 billion in 2025), commits 34.3 billion, merged PRs 1.3 billion and Actions jobs 45.4 billion, with the surge starting around June as agents went persistent and moved to the cloud.
• Agent PRs touch multiple files 87% of the time versus 79%, and 90% merged within two weeks by September, up from 83% in February.
• Paid Copilot developers used 47 distinct models in September, no model above 25% of requests, split 38/37/25 across general, reasoning and fast models; 55% of Copilot users now use agentic features.
• TypeScript stays first, Rust doubled, and detected AI involvement in merged PRs runs 44.1% for TypeScript, 41.1% Python, 40.4% Rust and 21.2% Java.
• GitHub counts 225 million developers, with the US at 38.3 million and India at 32.0 million; the agent count is a floor, the window changed to include private repos, and a push measures uploads, not code.

Sources: GitHub Blog: Octoverse 2026, Octoverse 2026 report page, The Next Web: GitHub Universe and Octoverse, Simon Willison: GitHub Octoverse 2026, Digit: Octoverse 2026 India vs US