🔍 Read the full analysis: 4 In 10 GitHub Pull Requests Start With AI Agents, Putting Review In Focus on ThorstenMeyerAI.com
Get business pricing on tech for your team
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
TL;DR
GitHub 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 reported that 90% of agent-started requests were merged within two weeks, but a merge does not establish that code was correct. GitHub’s new review tools put human oversight, approval rules and review costs at the center of its latest product changes.
More than 40% of GitHub pull requests in September began with a first draft from a coding agent the company could detect, up from 17% in March, GitHub said in its 2026 Octoverse report. The company counted 110 million agent-started pull requests in September, a shift that puts the capacity to review and approve code under new pressure.
GitHub defines an agent-started pull request as one whose first draft came from a coding agent the company can detect. It describes the resulting share as a floor, because its telemetry may miss some agent use. The measure includes public and private repositories unless otherwise specified, but GitHub did not publish a split between them for this figure. A request still counts as agent-started if a person later revises much of the draft.
The report says 90% of agent-started pull requests were merged within two weeks in September, compared with 83% in February. Pull requests without a detected agent held at about 86%. Those figures measure merging, not correctness, quality or the amount of follow-up work. Agent-started requests also more often changed multiple files: 87% affected more than one file, compared with 79% of requests without a detected agent.
Alongside the report, GitHub announced products aimed at review. Copilot code review Max uses several AI reviewers from different model providers and combines their feedback. GitHub also said Copilot review can count toward required approvals and introduced Copilot Fusion, a model router that selects models for different steps. 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 Becomes the New Constraint
The figures point to a change in where software work can slow down. If agents produce more first drafts quickly, teams still need people or review systems to check the changes, judge their fit and decide whether they should ship. GitHub’s product announcements reflect that concern: its focus was not only on generating code, but on review, routing and approval.
For engineering teams, the headline rate alone cannot show whether agent use is improving software. The 90% two-week merge rate may indicate that agent requests move through review, but merging is not proof of correctness. Organizations will need to track rework, reversions and incidents alongside request counts, and plan for the costs of both AI reviews and reviewer time.
The change also affects developers’ daily work. Writing a clear task for an agent and reading a multi-file diff critically become more central as agents handle more initial implementation. Junior developers may also have fewer small, routine tasks available for practice if those tasks are assigned to agents; the report does not measure that effect, but it is a concern raised by the shift in workflow.
Products Worth Considering
Bose QuietComfort Wireless Noise-Cancelling Over-Ear Headphones, Black
Bose QuietComfort Wireless Bluetooth Headphones with Active Noise Cancellation, Mic, USB‑C Charging, Deep Bass, 24‑Hour Battery – White Smok
Logitech MX Mechanical Wireless Illuminated Keyboard Tactile (Graphite)
As an affiliate, we earn on qualifying purchases.
How GitHub Measured Agent Requests
The Octoverse report is GitHub’s annual review of activity on its platform. The 2026 edition covers October 1, 2025, through September 30, 2026. GitHub says pull requests were counted when its systems detected an agent as the source of the first draft. That wording matters: the reported metric is about how a request began, not how much of its final code an agent wrote.
GitHub’s report uses several related labels, including “agent-started,” “agent-generated” and “agent-opened.” Its definition of the headline measure is the first-draft wording, so that is the most precise description. GitHub also recalculated its 2024 and 2025 figures using revised definitions and a revised time window, making comparisons with older Octoverse reports unsuitable.
The agent figures sit within a broader increase in platform activity. GitHub reported 1.3 billion merged pull requests during the year ending in September, up from 538 million in 2025 and 402 million in 2024. It also said 55% of Copilot monthly active users used agent features, compared with 35% a year earlier. These platform-wide measures offer scale, but do not by themselves establish what caused changes in productivity or code quality.
“A pull request whose first draft came from a coding agent that we can detect.”
— GitHub, in its Octoverse report
Limits of the September Count
The actual share of agent use is unknown because GitHub’s detection can miss agents. The company has not disclosed how much use goes undetected, nor provided a public-versus-private repository breakdown for the headline figure. Because the count is based on the first draft, it also does not indicate what portion of the final code remained agent-written after human edits.
The reported merge rates do not show whether merged changes were correct, needed later repair or caused incidents. GitHub’s report, as described in the source material, does not supply those outcome measures for agent-started requests. The effects on reviewer workloads and junior developers are also not quantified. GitHub’s own figures describe activity on its platform and should not be treated as a measure of all software development.
Teams Face Review Decisions
GitHub’s next step is putting its announced review products and approval options into use, with estimated per-review prices and separate Actions charges for teams to account for. The company has not supplied a future milestone in the source material for when it will publish a more detailed breakdown of agent detection or code outcomes.
For engineering organizations, the immediate decisions include where AI-generated review may count toward required approvals, how much review capacity to budget and which measures will show whether the workflow is working. The evidence needed to judge the shift goes beyond how many pull requests agents start: teams will need to examine review time, rework, reversions and production incidents as more agent-written changes pass through their processes.
Key Questions
What does GitHub mean by an agent-started pull request?
GitHub counts a pull request when its first draft came from a coding agent the company can detect. A person may later rewrite or revise part of that draft and the request still counts.
Does the 40% figure include every AI-written pull request?
No. GitHub says its detection may miss some agent use and calls the figure a floor. It has not quantified the amount of undetected use.
Does a high merge rate mean the code was correct?
No. GitHub reported that 90% of agent-started pull requests were merged within two weeks in September, but merging does not establish correctness or show whether changes required later repair.
How much does GitHub estimate an AI code review costs?
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 agent-written code that ships?
GitHub says developers remain accountable for shipped code. Its source material also says the cloud agent cannot approve or merge its own pull request, and the person who requested a change cannot approve it either.
Source: ThorstenMeyerAI.com
