🔍 Read the full analysis: How Teams Can Rethink GitHub Review As AI Agents Start 4 In 10 PRs on ThorstenMeyerAI.com
Get business pricing on monitors, keyboards and dev gear
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
TL;DR
GitHub’s Octoverse 2026 report says more than 40% of pull requests on the platform in September began with a detectable coding agent, up from 17% in March. The shift is prompting new review products, but GitHub’s figures do not show whether merged agent-written code is correct or how its use divides between public and private repositories.
More than 40% of pull requests on GitHub in September started with a coding agent’s first draft, up from 17% in March, according to the company’s Octoverse 2026 report. GitHub says that amounted to 110 million agent-started pull requests in one month, a change that makes code review and approval a more immediate constraint for teams relying on agents to write software.
GitHub defines an agent-started pull request as one whose first draft came from a coding agent the company can detect. The company says the measure is a floor, because some agent use may not be detected. A pull request still counts as agent-started even if a person later rewrites part of the draft. The report’s methodology says its metrics include public and private repositories unless otherwise noted; GitHub did not provide a public-versus-private breakdown for this figure.
The report also says 90% of agent pull requests were merged within two weeks in September, compared with 83% in February. Pull requests without a detected agent remained at about 86%. Those rates describe whether changes were merged, not whether they were correct, safe, or free of later rework. Agent-started changes also touched more files: 87% affected more than one file, compared with 79% of pull requests without a detected agent.
At its October 28 Universe conference, GitHub announced tools focused on review and workflow. Copilot code review Max runs several AI reviewers using different model providers and combines their feedback. GitHub also said Copilot review approvals can count toward required approvals, and introduced Copilot Fusion, a router that chooses models for different steps. The company estimates an average review costs $0.25 in Lite, $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 Becomes the New Constraint
When agents produce more proposed changes, teams must still decide which changes deserve attention and whether they are ready to ship. The central operational question is no longer just how quickly code can be drafted; it is whether reviewers have the time, context and authority to check it. Larger, multi-file changes can demand more review effort, though the report does not quantify how much extra time they take.
GitHub’s review pricing also makes the process a budget decision. Organizations may need to plan for review credits, Actions usage and human reviewer capacity, rather than treating agent output as cost-free. Counting pull requests alone could reward volume without showing whether changes reduce rework or introduce defects. Useful measures would include reversions, incidents and time spent correcting agent-written code, but the report does not provide those outcome comparisons.
For developers, the shift raises the value of defining a task precisely and inspecting a diff critically. For teams bringing in junior staff, it may also change how people learn: small, contained coding tasks have often served as early practice, while such tasks are among those agents can now draft. The extent of that effect on hiring and training remains unknown.
Products Worth Considering
LG UltraGear 32GX850A-B 32-Inch 4K UHD Glossy OLED Gaming Monitor
LG 34GX90SA-W 34-inch Ultragear WQHD OLED Curved Gaming Monitor
Samsung 27-Inch Odyssey QD-OLED G8 4K 240Hz Gaming Monitor
As an affiliate, we earn on qualifying purchases.
How GitHub Defines Agent-Started Work
The Octoverse 2026 report covers October 1, 2025, through September 30, 2026. GitHub recalculated its 2024 and 2025 figures using updated definitions and a new reporting window, so older Octoverse totals are not directly comparable. The report uses several nearby labels, including “agent-generated” and “agent-authored,” but its definition and headline describe work that started with an agent.
The wider platform also grew during the reporting period. GitHub reported 1.3 billion merged pull requests, up from 538 million in 2025 and 402 million in 2024. It recorded 17.9 billion code pushes and 34.3 billion commits over the 12 months to September. These platform-wide figures provide scale, but they do not establish that agents alone caused the increase.
GitHub says 55% of monthly active Copilot users now use agent features, up from 35% a year earlier, and that use of its cloud agent grew 230% year over year. Those are company-reported product figures. They indicate wider adoption, but do not independently measure code quality or business results.
“A pull request whose first draft came from a coding agent that we can detect.”
— GitHub, in the Octoverse 2026 report
Quality and Accountability Still Need Testing
The report does not establish whether the rise in merged agent pull requests has improved software quality. Merging is not proof of correctness, and GitHub did not report comparable rates for defects, security problems, reverts or rework in the supplied figures. Nor did it publish a breakdown showing how much of the agent-started activity came from public projects versus company repositories.
Detection limits mean the 40%-plus figure is not a full census of agent-written work. The report’s terminology also varies across text and charts, so comparisons should use its defined “agent-started” measure rather than treating every label as interchangeable. GitHub says the person who requested a change cannot approve it and that its cloud agent cannot approve or merge its own pull request. How organizations will configure approval policies, and how much weight they will give AI-generated review, depends on their own controls.
Teams Must Set Review and Cost Rules
GitHub’s newly announced review features put practical choices in front of engineering teams: which changes merit multi-model review, when an AI review may count toward a required approval, and how to reserve human review for changes with greater risk. The company’s published estimates give teams starting points for budgeting, but actual spending will depend on usage and the additional cost of Actions minutes.
The next meaningful test will be operational rather than a change in the headline share: whether teams can maintain review quality as agent output grows. Companies will need to compare review costs and time with outcomes such as corrections, reversions and incidents. GitHub has not supplied those results in the report, so the impact of its new review tools remains to be measured.
Key Questions
What does GitHub mean by an agent-started pull request?
It means the first draft of the proposed code change came from a coding agent that GitHub can detect. A person may later edit the draft, and it still counts under that definition.
Did agents start more than 40% of pull requests in every repository?
No. GitHub reported an overall platform figure for September and did not publish a public-versus-private split in the supplied report. Its detection may also miss some agent use.
Does the 90% merge rate show that agent-written code is reliable?
No. The figure measures the share merged within two weeks, not whether the code was correct or whether it later needed repair. GitHub did not provide comparable quality or incident data in these figures.
How much does GitHub say an AI 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.
Who is responsible for code that an agent proposes?
GitHub says developers remain accountable for code that ships. It also says its cloud agent cannot approve or merge its own pull requests, and the person who requested a change cannot approve it.
Source: ThorstenMeyerAI.com
