🔍 Read the full analysis: GitHub Pull Requests: AI Agents Start 4 In 10, And Review Takes Center Stage 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’s Octoverse report says more than 40% of pull requests on the platform in September 2026 started with a detectable coding agent, up from 17% in March. GitHub introduced new review and routing tools as agent-written changes grow, but says developers remain accountable for code that ships.
More than four in 10 GitHub pull requests started with a coding agent in September, according to the platform’s 2026 Octoverse report, up from 17% in March. GitHub says the increase is shifting attention from producing a first draft to reviewing and approving code, and it announced new tools for that work at its October 28 Universe conference.
GitHub counted 110 million agent-started pull requests in September. It defines one as a pull request whose first draft came from a coding agent the company can detect. GitHub calls its count a floor because it may miss some agent use; a pull request still qualifies if a person later rewrites part of the draft. The reported share includes public and private repositories, with no separate breakdown published for the two.
In September, 90% of agent-started pull requests were merged within two weeks, compared with 83% in February. The merge rate for pull requests without a detected agent stayed at about 86%. A merge records that a change was accepted into a codebase; it does not establish that the code was correct, safe or free of later rework.
GitHub’s new offerings put review at the center. Copilot code review Max brings together reviews from multiple AI model providers into one combined review. GitHub also made Copilot approvals eligible to count toward required approvals and released Copilot Fusion, a system that selects models for different steps. GitHub estimates a review costs $0.25 in Lite, $1.50 in Balanced and $10 in Max, plus GitHub Actions minutes.
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 Becomes the New Bottleneck
The rise in agent-started changes matters because a faster first draft can shift work rather than remove it. Each proposed change still needs someone to understand its purpose, inspect the diff, check how it fits the codebase and decide whether it should ship. GitHub’s new review products and per-review estimates reflect that growing workload, though the company has not shown that AI review reduces total review time or catches defects at a particular rate.
For engineering teams, the reported scale makes review capacity a practical planning issue alongside computing costs. Teams will need to decide where machine-generated reviews can help, how much human review is required and whether an AI approval can satisfy a repository’s rules. Merge counts alone cannot show whether the change was useful: rework, reversions and production incidents are also relevant measures.
The shift may affect how developers learn and work, too. Small, well-scoped tasks have often given less experienced engineers a way to gain familiarity with a codebase. If agents take on more of that first-draft work, teams may need to be deliberate about how junior developers build skills in reading, testing and reviewing code. That is a potential consequence, not an outcome established by the report.
Products Worth Considering
Logitech MX Keys S Combo Wireless Keyboard and Mouse, Graphite
Logitech MX Ergo S Advanced Wireless Trackball Mouse USB-C – Graphite
Logitech MX Master 4 Ergonomic Wireless Mouse with Haptic Feedback, Graphite
As an affiliate, we earn on qualifying purchases.
How GitHub Counts Agent Changes
The figures appear in GitHub’s annual Octoverse report, which covers October 1, 2025, through September 30, 2026. GitHub says the agent measure is based on detectable signals, not a complete audit of who or what wrote every line. It describes the counted pull requests using several related terms, including agent-started, agent-generated and agent-authored; its definition points most precisely to a change whose first draft came from an agent.
GitHub reported broader growth in activity during the same period: 1.3 billion merged pull requests, compared with 538 million in 2025 and 402 million in 2024. The company also recalculated 2024 and 2025 figures under updated definitions and reporting windows, so those historical numbers are not directly comparable with earlier Octoverse reports.
The report says 87% of agent-started pull requests changed more than one file, versus 79% of pull requests without a detected agent. That difference suggests reviewers may need to follow changes across more parts of a project, but file count by itself does not establish complexity or quality. GitHub also says 55% of Copilot monthly active users now use agent features, up from 35% a year earlier.
“A pull request whose first draft came from a coding agent that we can detect.”
— GitHub, in the Octoverse report definition
Limits of the September Count
The 40%-plus share depends on signals GitHub can detect, and the company has not published a measure of missed agent activity. It also has not supplied a public-versus-private repository split for this statistic. Because GitHub’s report uses several labels for agent-related pull requests, readers should distinguish its specific first-draft definition from broader claims that every counted change was written entirely by AI.
The reported merge rate is not a quality measure. The material provided does not state how often agent-started changes were later revised, reverted or linked to incidents, nor does it give a controlled comparison showing whether AI review improves outcomes. GitHub’s per-review prices are estimates and do not include GitHub Actions minutes, so a team’s full costs will vary.
GitHub says its cloud agent cannot approve or merge its own pull requests, and the person who requested a change cannot approve it either. Details about how organizations will configure Copilot approvals, and how those approvals perform in practice, are not established by the report. The extent to which agent use varies by company, language or project type also remains unclear.
Teams Set Review Rules
The immediate next step is for companies using GitHub’s tools to set policies for when AI reviews can count toward required approvals and how much human inspection each change needs. Teams can compare the estimated review tiers with their own workloads, while accounting for Actions minutes and the cost of follow-up work. GitHub’s announced tools provide new options; the report does not establish how widely they will be adopted or whether they will reduce bottlenecks.
As use grows, a useful test will be whether teams track outcomes beyond how many pull requests agents start or how many are merged. Measures such as rework, reversions and production incidents can help teams judge whether the new workflow is working for their codebases. GitHub has not yet provided those results as part of the figures cited here.
Further comparisons will depend on future reporting and on consistent definitions of agent-started work. For now, GitHub’s September figures show a major change in how proposed code begins on its platform, while the effects on software quality, reviewer workload and developer training remain open questions.
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 the company can detect. A person may edit the draft afterward and it can still count.
How many GitHub pull requests started with agents in September?
GitHub reported 110 million agent-started pull requests, representing more than 40% of pull requests that month. The company says its count is a floor because some agent use may go undetected.
Does a high merge rate mean agent-written code is correct?
No. GitHub reported that 90% of agent-started pull requests were merged within two weeks, but being merged is not the same as being correct or free of later rework and defects.
How much does GitHub estimate an AI code review costs?
GitHub estimates $0.25 for Lite, $1.50 for Balanced and $10 for Max per review, plus GitHub Actions minutes. Actual total costs may vary.
Who is responsible for code an agent helps create?
GitHub says developers remain accountable for what ships. Its cloud agent cannot approve or merge its own pull request, and the person who requested the change cannot approve it either.
Source: ThorstenMeyerAI.com
