🔍 Read the full analysis: 4 In 10 GitHub Pull Requests Begin With AI—Review Work Is Changing on ThorstenMeyerAI.com
Get hardware and tech essentials delivered free with Prime
- Fast, free delivery on millions of items
- Prime Video, Amazon Music and more included
- Member-only deals all year
TL;DR
GitHub says more than 40% of pull requests on its platform in September began with a detectable coding agent, up from 17% in March. The company has introduced new review and routing tools as the volume of agent-written code grows, but a merged pull request is not proof that the code is correct.
More than 40% of GitHub pull requests in September began with a first draft from a detectable coding agent, up from 17% in March, according to the company’s 2026 Octoverse report. GitHub says that amounts to 110 million agent-started pull requests in September, a shift that is putting more emphasis on reviewing and approving code rather than producing its first draft.
GitHub released the report on October 28 at its Universe conference. It defines an agent-started pull request as one whose first draft came from a coding agent the company can detect. GitHub says its count is a floor, not a complete tally, because some agent use may not be detected. A pull request still counts as agent-started even if a person later changes much of the draft.
The figures cover public and private repositories, according to the report’s methodology, but GitHub does not provide a separate breakdown for the two. It also says 90% of detected agent-started pull requests in September were merged within two weeks, compared with 83% in February. The corresponding rate for pull requests without a detected agent was about 86%. Being merged records an approval outcome; it does not establish that code is correct or free of problems.
GitHub’s new products focus on oversight. The company introduced Copilot code review Max, which draws on several AI reviewers from different model providers and combines their feedback, and Copilot Fusion, a system that routes tasks to models. It also made Copilot approvals eligible to count toward required approvals. GitHub estimates an average review costs $0.25 for Lite, $1.50 for Balanced and $10 for Max, with GitHub Actions minutes charged separately.
Infographic 01 · The agent share
Four in ten pull requests now start with an agent.
GitHub’s own telemetry, March to September 2026. The first draft comes from an agent; review and merge stay with people.
Merged within two weeks
Share of pull requests merged within 14 days
Merged fast is not the same as correct. GitHub publishes no revert or defect rate.
Bigger changes to review
Share of pull requests that change more than one file
More files per change means more for a reviewer to check.
What GitHub counts: an “agent-started pull request” is one “whose first draft came from a coding agent that we can detect.” Undetected agents are missing, so GitHub calls the figure a floor. Metrics cover public and private repositories unless noted.
Source: GitHub Octoverse 2026, 28 Oct 2026 (text, chart captions, glossary and methodology). All figures are GitHub’s own.
Powered by Thorsten Meyer AI
Review Is Becoming the Constraint
The figures point to a change in where software teams may spend time. If agents create first drafts faster, human review capacity can become a limiting step: reviewers must understand changes, check whether they fit the task and decide whether to approve them. GitHub’s product announcements reflect that pressure, but the company’s adoption numbers do not by themselves show that review is taking longer or that code quality has worsened.
For organizations, the reported per-review prices make automated review a budget decision as well as a workflow choice. Teams will need to decide where AI feedback is useful, what approvals must remain human, and how to judge results. Pull-request volume and merge rates alone do not measure rework, bugs, reversions or incidents. Those outcomes would help show whether increased agent use is improving delivery or shifting effort into checking and fixing code.
The shift may also affect how developers gain experience. The source argues that small starter tasks often used by junior developers are among the work agents can take on. That is a concern about training and job design, not a finding established by the Octoverse figures. Teams adopting agents will still need ways for less-experienced developers to learn how to read code and build sound judgment.
Products Worth Considering
LG UltraGear 45GX900A-B 45-Inch Curved OLED Gaming Monitor
LG 24G414B-B 24-Inch UltraGear Full HD IPS Gaming Monitor
LG UltraGear 32GX850A-B 32-Inch 4K UHD Glossy OLED Gaming Monitor
As an affiliate, we earn on qualifying purchases.
How GitHub Counts Agent Work
The Octoverse report covers October 1, 2025, through September 30, 2026. GitHub says its 2024 and 2025 figures were recalculated using the updated definitions and reporting window, so older Octoverse totals are not directly comparable. The report’s wording varies between “agent-started,” “agent-generated” and “agent-authored”; the most specific definition is a pull request whose first draft came from a detectable agent.
Other activity totals help explain the scale of the change, though they do not isolate its effects. GitHub reports 1.3 billion merged pull requests during the year, compared with 538 million in 2025 and 402 million in 2024. It also says 87% of detected agent pull requests changed more than one file, against 79% of other pull requests. Larger, multi-file changes can ask more of reviewers, but the report does not quantify review time or error rates for those changes.
GitHub reports that 55% of Copilot monthly active users use agent features, up from 35% a year earlier, and that use of its cloud agent grew 230% year over year. These are company-reported platform measures. They describe adoption, not an independent assessment of productivity or software quality.
“A floor, since not all agent use may be detected.”
— GitHub’s Octoverse report
What the Adoption Figures Cannot Show
The report does not publish an open-source versus private-repository split for agent-started pull requests, nor does it identify how many agent-assisted drafts go undetected. Because the count includes any pull request whose first draft came from a detectable agent, it does not reveal how much code people later rewrote or how much of the final change was authored by a person.
The reported merge rate also leaves important questions unanswered. GitHub’s figures do not say whether agent-started changes produced more defects, required more revisions, or caused more incidents after release. The source material provides no independent verification of those outcomes. It is also not clear how frequently organizations will count Copilot approvals toward required reviews or how much the quoted review estimates will vary with actual use and Actions minutes.
Teams Must Track Review Outcomes
GitHub’s announced review and routing tools are now the company’s response to the growth in agent-written drafts. The next useful evidence will be whether teams adopt them and how they affect review time, rework and post-release defects—measures not established by the adoption figures in this report. GitHub has not provided an independent evaluation of those effects in the supplied material.
Organizations using coding agents will have to set their own approval policies, decide which changes require human sign-off and account for review credits alongside other computing costs. GitHub says a developer remains accountable for shipped code and that its cloud agent cannot approve or merge its own pull request; it also says the person who requested a change cannot approve it. How those safeguards work across different company workflows remains to be seen.
Key Questions
What does GitHub mean by an agent-started pull request?
It is a pull request whose first draft came from a coding agent GitHub can detect. GitHub says some agent use may go undetected, so its total is a floor. Human edits made after the first draft do not change the classification.
Does the report say that AI began most GitHub pull requests?
No. GitHub says more than 40% in September began with a detectable agent, up from 17% in March. That is a large share, but below half.
Does a high merge rate prove agent-written code is good?
No. GitHub reports that 90% of detected agent-started pull requests in September were merged within two weeks, but merging is not a measure of correctness. The report does not establish comparative defect, rework or incident rates.
How much does GitHub say an AI code review costs?
GitHub estimates an average Copilot review at $0.25 for Lite, $1.50 for Balanced and $10 for Max. GitHub Actions minutes are additional, and actual spending depends on use.
Who is responsible for code produced by an agent?
GitHub says developers remain accountable for what ships. It says its cloud agent cannot approve or merge its own pull requests, and the person who requested the change cannot approve it either.
Source: ThorstenMeyerAI.com
