🔍 Read the full analysis: AI Agents Start More GitHub Pull Requests, Shifting Attention To Review on ThorstenMeyerAI.com
Prime made for students and young adults
- Fast, free delivery for dorm and study essentials
- Prime Video and Amazon Music included
- Member-only deals
TL;DR
GitHub’s Octoverse report says more than 40% of pull requests on its platform in September 2026 began with a detectable coding agent, up from 17% in March. The company also announced review and model-routing tools as it focuses on checking agent-written code; its counts do not establish that merged changes are correct.
More than 40% of pull requests on GitHub in September started with a detectable coding agent, up from 17% in March, according to the company’s Octoverse report released October 28. The shift puts pressure on review and approval: GitHub announced new tools for checking agent-written changes, while stressing that people remain accountable for code that ships.
GitHub says the September figure represents 110 million agent-started pull requests in one month, about six times the March volume. It defines an agent-started pull request as one whose first draft came from a coding agent the platform can detect. The company describes its tally as a floor because some agent use may go undetected. A pull request is a proposed code change that is reviewed before it is merged into a project.
The report says 90% of pull requests with a detected agent were merged within two weeks in September, compared with 83% in February. The rate for pull requests without a detected agent remained around 86%. Those figures describe merge rates, not whether the changes were accurate, safe or useful. GitHub also reports that 87% of agent-started pull requests changed more than one file, compared with 79% of those without a detected agent.
At its October 28 announcements, GitHub introduced Copilot code review Max, which uses reviewers from multiple model providers and combines their feedback, and Copilot Fusion, which routes tasks to models. It also made Copilot review approvals eligible to count toward required approvals. GitHub estimates an average review costs $0.25 in Lite mode, $1.50 in Balanced and $10 in 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 Work Is Becoming the Constraint
The data points to a change in where development teams may spend time: agents can produce an initial code change quickly, but people still need to understand the diff, test its effects and decide whether it should be merged. With agent pull requests rising and many touching multiple files, review capacity may become a practical limit on how much generated code teams can responsibly accept.
That has operational and budget implications. GitHub’s per-review estimates make AI review a metered expense alongside compute, while human review time remains necessary. Teams will need to decide whether automated approvals can satisfy particular rules and how to evaluate quality through outcomes such as rework, reverts and incidents—not just the number of pull requests or their merge rate.
The shift may also affect how developers learn. Routine, small coding tasks can provide early-career practice, and the source report argues that such tasks are among those agents can take on. The report does not quantify the effect on hiring or training, but it raises a practical question for teams: how will junior developers gain experience in writing, testing and reviewing code when agents handle more first drafts?
Popular right now
Ninja Foodi 10-Quart DualZone 6-in-1 Air Fryer, Black, DZ302
Ninja CREAMi 7-in-1 Ice Cream Maker, NC301
Breville BES870XL Barista Express Espresso Machine, Brushed Stainless Steel
As an affiliate, we earn on qualifying purchases.
How GitHub Defines Agent Work
GitHub’s 2026 Octoverse report covers the year from October 1, 2025, to September 30, 2026. Its agent measure is based on signals the platform can observe. A pull request can still count as agent-started even if a person later revises a substantial portion of the draft. The report’s wording varies between “agent-started,” “agent-generated” and “agent-authored”; “agent-started” most closely reflects the stated definition.
The report’s methodology says figures generally include public and private repositories unless a metric is marked otherwise. GitHub does not provide a public-versus-private breakdown for the agent share. It also recalculated 2024 and 2025 figures under updated definitions and reporting windows, so comparisons with earlier Octoverse reports may not be direct.
The rise coincides with broader growth on the platform. GitHub reports 1.3 billion merged pull requests during the 12 months to September, compared with 538 million in 2025 and 402 million in 2024. It also says 55% of Copilot’s monthly active users used agent features, up from 35% a year earlier, and that use of its cloud agent rose 230% year over year. These are figures reported by GitHub, not independent measurements.
“A pull request whose first draft came from a coding agent that we can detect.”
— GitHub, in its Octoverse report
What the Counts Cannot Show
The report does not show how many agent-started changes contained defects, needed substantial rework, or caused incidents after merging. A 90% merge rate is not a measure of correctness, and the figures do not establish whether agents improve overall productivity or code quality. GitHub’s data also lacks a breakdown of agent-started pull requests by public and private repository.
The measurement depends on detectable agent signals, so the actual share may be higher, but the source does not quantify how much higher. Nor does the report clarify in detail how much human rewriting can occur while a pull request remains classified as agent-started. Copilot approvals counting toward required approvals also leave questions for individual organizations about which policies and changes should permit that form of review.
Teams Set Review Rules and Budgets
GitHub has released the review and routing tools alongside its report, but the source material gives no date for a further rollout milestone or independent evaluation of their performance. Teams adopting them will need to set review policies, choose when an AI approval can count and account for the published per-review estimates and associated Actions usage.
The next useful evidence will be whether organizations can track outcomes beyond throughput: defects, reverts, review time and the amount of human rework on agent-generated changes. GitHub’s figures establish that agent-started pull requests have become a large share of platform activity; whether that activity translates into better or faster software remains to be measured.
Key Questions
What does GitHub mean by an agent-started pull request?
GitHub defines it as a pull request whose first draft came from a coding agent it can detect. A person may later edit the draft and it can still count under that definition.
Does the 90% merge rate mean agent-written code is correct?
No. The figure measures how many detected agent pull requests were merged within two weeks. It does not report correctness, safety or post-merge performance.
How much does Copilot code review cost?
GitHub estimates an average review at $0.25 for Lite, $1.50 for Balanced and $10 for Max. GitHub Actions minutes are additional.
Who is responsible for code an agent submits?
GitHub says developers remain accountable for shipped code. Its cloud agent cannot approve or merge its own pull requests, and the person who requested a change cannot approve it either.
Yes. GitHub calls its figure a floor because it cannot detect every use of an agent. The report does not estimate the scale of that undercount.
Source: ThorstenMeyerAI.com
