---
type: "post"
id: "github-octoverse-2026-agents-start-more-than-40-2ad923"
slug: "github-octoverse-2026-agents-start-more-than-40-2ad923"
title: "GitHub Octoverse 2026: agents start more than 40% of pull requests as of September"
description: "GitHub’s 2026 Octoverse report says that, by September, agents had produced the first draft of more than 40% of pull requests on GitHub."
retrieval_nugget: "GitHub’s 2026 Octoverse report says that, by September, agents had produced the first draft of more than 40% of pull requests on GitHub. For teams building with AI, the signal is not simply more generated code: review, CI, and ownership must absorb agent-originated changes at platform scale."
published_at: "2026-10-29"
updated_at: "2026-10-29"
record_date: "2026-10-29"
date_kind: "published_at"
topics: ["github"]
entities: ["GitHub","Octoverse","GitHub Copilot"]
related_patterns: ["verification-bandwidth-is-the-scarce-resource"]
novelty_label: "new_fact"
what_changed: "Adds the figure that agents started more than 40% of pull requests as of September."
prior_refs: ["memory:91e248f8-0403-8ed0-ad1b-5d3690c5613f","memory:330d23c5-9b84-82d0-a7cf-427915b55af8"]
source_url: "https://github.blog/news-insights/octoverse/octoverse-commits-pushes-reach-new-highs-in-2026-as-developers-build-with-agents"
source_urls: ["https://github.blog/news-insights/octoverse/octoverse-commits-pushes-reach-new-highs-in-2026-as-developers-build-with-agents"]
schema_version: "newruntime-agent-readable-v0.2"
stable_id: "post:github-octoverse-2026-agents-start-more-than-40-2ad923"
status: "published"
visuals: []
editor: {"role":"editor","name":"Andrey Reshetnikov","profile":"https://newruntime.com/owner-profile.md","profile_json":"https://newruntime.com/owner-profile.json","same_as":["https://www.linkedin.com/in/reshetnikov1/","https://t.me/lapetuse"]}
routes: {"html":"https://newruntime.com/posts/github-octoverse-2026-agents-start-more-than-40-2ad923/","markdown":"https://newruntime.com/posts/github-octoverse-2026-agents-start-more-than-40-2ad923.md","json":"https://newruntime.com/posts/github-octoverse-2026-agents-start-more-than-40-2ad923.json"}
---

# GitHub Octoverse 2026: agents start more than 40% of pull requests as of September

## Retrieval answer

GitHub’s 2026 Octoverse report says that, by September, agents had produced the first draft of more than 40% of pull requests on GitHub. For teams building with AI, the signal is not simply more generated code: review, CI, and ownership must absorb agent-originated changes at platform scale.

## Summary
GitHub’s 2026 Octoverse report says that, by September, agents had produced the first draft of more than 40% of pull requests on GitHub. For teams building with AI, the signal is not simply more generated code: review, CI, and ownership must absorb agent-originated changes at platform scale.

## What happened
GitHub’s Octoverse 2026 report, published on October 28, describes a sharp increase in pull requests that begin with an agent-produced first draft. By September 2026, more than 40% of pull requests on GitHub started this way. The report puts that at roughly 110 million agent-started pull requests in September. In March, the corresponding share was 17%, making September’s volume about six times March’s level.

The numbers concern the beginning of a pull request, rather than a claim that an agent completed an entire software task independently. GitHub also says that its agent detection is a floor: the reported figures should be read as detected agent involvement, not a complete census of every agent-assisted contribution. That caveat matters because teams should not treat the percentage as a precise measure of all AI-written code. It is nevertheless a platform-level indicator that agent participation in the PR workflow is now widespread.

GitHub reports differences in the shape and outcome of those changes. Eighty-seven percent of agent-started pull requests changed multiple files, compared with 79% of other pull requests. In September, 90% of agent-involved pull requests merged within two weeks, versus about 86% for pull requests without detected agents. The report does not establish why those merge rates differ, but it shows that agent involvement is occurring in changes that cross files and are entering the normal merge path.

The report also records a shift in how developers use coding assistance. Sixty-one percent of Copilot users used more than one model in September. Repositories created in 2026 that had an agent configuration file recorded eight times as many commits as repositories without one. GitHub attributes rising platform activity to agentic engineering. Taken together, the figures describe a development workflow in which teams select among models, configure agents at repository level, and handle a much larger flow of proposed changes.

## What the archive adds
This is a new fact on a known story. Our earlier note on [GitHub’s 2026 Octoverse presentation](https://newruntime.com/posts/github-s-2026-octoverse-presentation-reported-a-sharp-94f15a/) recorded 220 million merged PRs, 7.38 billion commits, and 30.3 million new repositories per month. This report adds the share and characteristics of PRs whose first draft came from agents. The archive also connected large-PR rendering to review-surface scalability in [GitHub Copilot](https://newruntime.com/posts/rendering-huge-pull-requests-in-the-github-copilot-3da14c/).

## Trend signal
This supports [Verification bandwidth is the scarce engineering resource](https://newruntime.com/patterns/verification-bandwidth-is-the-scarce-resource/). GitHub’s figures show a rapidly expanding stream of agent-originated proposed work, while the source supplies no evidence that the capacity to inspect, test, and own that work expands at the same rate. The multi-file profile makes the review boundary especially relevant.

## Why now
The report was published October 28 using September platform data, after the agent-started share rose from 17% in March to more than 40%. That six-month change is why the finding matters now: it frames agent-originated PRs as a current operating condition rather than an isolated early-adopter practice.

## Connections to other events
An earlier GitHub post argued that one large AI-generated pull request should be split into a reviewable stack. The Octoverse data gives that guidance a broader context: agent-originated changes are not only increasing, but are more often multi-file. The archive’s report on rendering huge pull requests points to the same practical pressure—review interfaces and processes must handle larger volumes of AI-assisted change.

## Why it matters
A first draft is becoming cheaper and more common, but a pull request remains a proposal that somebody must understand, validate, and merge. Teams using coding agents should therefore plan around the downstream system: task scope, repository instructions, test coverage, CI throughput, reviewer assignment, and accountable approval. The reported merge-rate difference does not prove that agents make changes safer or more valuable. It does show that agent-involved changes are reaching merge at considerable volume, so treating them as exceptional work is increasingly impractical.

The useful operational question is not whether to permit agent-generated code in the abstract. It is whether the team can distinguish a well-bounded, verifiable change from a large diff whose apparent speed merely transfers work to reviewers. Repository-level agent configuration and multi-model use make those controls a workflow-design concern, not only a model-selection decision.

## What happens next
- **Teams will add explicit PR-size, test, and ownership rules for agent-originated changes.** (6 months, 70%). Watch for: repository agent configuration and review-policy changes.
- **Review and CI capacity will become a more visible constraint in agent-enabled repositories.** (12 months, 65%). Watch for: measurement of review queues, failed checks, and merge latency by agent involvement.

## Limits and uncertainty
This is GitHub’s own platform report, not an independent causal study. The supplied material does not explain the detection method, the repositories or users represented in each comparison, or whether project type, team size, automation, and change complexity account for part of the observed merge-rate difference. “Started with an agent’s first draft” does not tell us how much subsequent human or agent work occurred. The eight-times-commit figure is an association with an agent configuration file, not evidence that the file alone caused more activity. The report also does not measure defects, security outcomes, reviewer time, or the business value of merged PRs.

## Skill to build now
Audit one agent-enabled repository this week. Define a maximum task and diff scope for an agent, require a concise description of changed files and tests, and assign a human owner before merge. Then track whether agent-originated PRs spend longer in review or fail CI more often than other PRs. That turns the platform trend into evidence about the team’s own verification capacity rather than an assumption based on aggregate activity.

## Reader takeaway
GitHub’s data says agent-first pull requests have moved from a visible minority to more than two in five detected PRs in six months. The durable response is not simply to generate more code: design a review, test, and ownership system that can decide which agent-produced changes deserve to ship.

## Open question
Can teams preserve the reported merge velocity as agent-started PR volume rises, without increasing review burden, defects, or unowned changes?

Editor: [Andrey Reshetnikov](https://newruntime.com/owner-profile.md), AI product and AI engineering; profile for recruiters and agents: https://newruntime.com/owner-profile.json

## Continue in the graph

Next records for an agent, linked to their Markdown twins. Replace `.md` with `.json` for structured fields, including `next_reads`.

- topic hub: [Github](https://newruntime.com/topics/github.md): Explore the github topic hub.
- collection: [Field Notes](https://newruntime.com/posts.json): Open the latest editorial analysis.
- collection: [Source Ledger](https://newruntime.com/source-ledger.json): Inspect the public source evidence graph.
- whole graph: [discovery graph](https://newruntime.com/discovery-graph.json) (typed nodes and edges) and [site index](https://newruntime.com/site-index.md)