AI agents now start more than 40% of GitHub pull requests. What the 2026 numbers show, and what they don't
Developers increasingly hand the first draft of a change to AI agents. The 2026 data shows how far that has gone, and where it stops.
Updated November 15, 2026

In short: By September 2026, developers started more than 40% of GitHub pull requests with an AI agent, up from 17% in March. Stack Overflow's 2026 survey finds 66% of AI users work in coding agents and 63% in chatbots. Neither report measures side projects.
Key Facts
- Pull requests started with an AI agent on GitHub, September 2026
- More than 40%
- As of September 30, 2026
- Pull requests started with an AI agent on GitHub, March 2026
- 17%
- As of March 31, 2026
- Agent-started pull requests on GitHub in September 2026
- 110 million
- As of September 30, 2026
- Respondents who use AI coding assistants or agents and use them daily
- 73%
- As of October 6, 2026
- Survey respondents who used AI at least an hour a day
- 80%
- As of October 6, 2026
How much of GitHub's work now starts with an AI agent?
Developers on GitHub increasingly ask an agent for the first draft of a change. GitHub reports that as of September 2026, more than 40% of pull requests started with an agent, up from 17% in March 2026 [1]. That is 110 million pull requests in one month, about six times March's volume [1].
GitHub describes this as a change in where work starts. In its words, for a growing share of work, the first move is now directing an agent, and review is how developers shape what comes back [1].
The number depends on how GitHub counts. It defines an agent-started pull request as one “whose first draft came from a coding agent that we can detect” [1]. The company also says the count is a floor: “It is a floor, since not all agent use may be detected” [1]. Read 40% as a minimum, not a ceiling.
The merge data points the same way. In September 2026, 90% of pull requests with agent involvement merged within two weeks, against about 86% of pull requests without detected agent involvement [1].
GitHub's report also links agent setup to activity. Among repositories created this year, the average repo with an agent configuration file logged eight times as many commits as the average new repo without one [1]. The comparison is observational. The post does not show that agent setup causes the difference.
Which AI tools do developers use most?
The Stack Overflow 2026 survey asked about AI use across the profession. Many AI users work in coding agents (66%) and chatbots (63%) [2]. The agents respondents name most are Claude Code (66%) and GitHub Copilot (59%) [2]. The post does not state the base for those two figures, so read them as the agents people name most, not as a share of all developers.
The same post says that when developers build with AI, they mostly generate code in familiar areas (67%) and debug (61%) [2].
How often do developers use AI?
Among respondents who use AI coding assistants or agents, 73% use them daily [2]. The official survey page says 80% of respondents used AI at least an hour a day [3].
The two figures use different bases. The first counts only people who use coding tools. The second counts all respondents. Keep them apart when you quote them, because a reader who sees both may assume they describe the same group.
Sentiment follows the same pattern. The survey reports that 62% of all respondents view AI at least somewhat positively, rising to 71% among daily users [2]. It also gives 93% as the share who say source attribution is required to trust what AI gives [2].
Do the numbers show developers use AI on side projects?
No. Neither the GitHub report nor the Stack Overflow survey measures side projects. GitHub's figures count activity on its platform [1], and the survey asks about AI use in general [2].
The survey does report one related figure. The share of respondents who worked in an organization of one rose from 4% to 10% [2]. That describes how people are employed, not what they build on the side. The post links the figure to anecdotes about people using AI agents to work solo, and it gives no figure for that use [2].
That gap matters most for anyone building alone. The figures show heavy agent use on GitHub and among survey respondents. They do not show whether agents help a solo developer finish a weekend project.
How should you read these numbers?
Three points matter before you repeat any figure.
- The bases differ. The 66% for coding agents is a share of AI users. The 40% for pull requests is a share of GitHub pull requests.
- Both sources are platform data. GitHub reports its own activity, and Stack Overflow reports its own survey. Attribute each figure to its source.
- Counts are floors or samples. GitHub's agent count misses agent use it cannot detect [1]. Survey percentages describe the people who answered the survey.
What can you do with this?
If you plan a side project with an AI agent, the data gives you one practical test. Track a few weeks of sessions and note which tasks you finished. That tells you more about your own project than an industry average does.
Neither source gives a figure for solo projects.
Frequently Asked Questions
What counts as an agent-started pull request on GitHub?
GitHub counts a pull request whose first draft came from a coding agent that it can detect. The count is a floor, because not all agent use is visible to GitHub.
Does the 2026 data show developers use AI on side projects?
No. Neither the GitHub report nor the Stack Overflow survey measures side projects.
Which coding agents do developers name most?
The Stack Overflow survey names Claude Code (66%) and GitHub Copilot (59%) as the agents respondents choose most. The survey post does not state the base for these figures.
How often do developers use AI?
Among respondents who use AI coding assistants or agents, 73% use them daily. The official survey page says 80% of respondents used AI at least an hour a day.
Sources
- 1.
- 2.
- 3.Stack Overflow Developer Survey 2026 (opens in new tab) — Stack Overflow Developer Survey
Tags
Written with AI assistance and reviewed by a human editor before publication.How we produce this content.


