AIdataNewsGlobal Read: What Octoverse 2026 Data Says About Agents, Model Choice and Builders Beyond Software
AI & Data News · General desk · By David Nakamura
A software-production dataset has moved into the small set of indicators that global AI observers actually watch, and the 2026 edition of Octoverse, published by GitHub, is the reference copy. It records code being pushed 17.9 billion times in the past 12 months, a total GitHub places above its own combined output for 2024 and 2025. The change underneath that figure is procedural rather than volumetric: an increasing share of work now opens with somebody instructing an agent, with the developer's judgment applied on the way back. Read as macro data, the report describes an AI build economy expanding in more countries, more sectors and more models at once.
Five findings that frame the reporting year
- Agent configuration tracks pace. Among repositories created this year, the average new repo carrying an agent configuration file logged eight times as many commits as the average new repo without one.
- Delegation has become the default opening move. By September, developers started more than 40% of pull requests with an agent, against 17% in March, which makes review the place where humans shape output.
- Portfolios beat singletons. In September, 61% of GitHub Copilot users ran more than one model, with requests spread across general purpose, deep reasoning and planning, and fast and lightweight tiers.
- Industry breadth is newly measurable. For the first time public repositories are viewed by industry, and repos serving mining, construction, health care, and agriculture have each grown several times over since 2022.
- Projections aged within twelve months. One year into last year's five-year developer growth projections, every country projected had already closed at least a fifth of its distance to 2030, and the forecasting models no longer agree on where things go next.
Throughput: the arithmetic behind the surge
Volume alone would not explain the year, so the report's automation and merge statistics matter. Commits across public and private repository types reached 34.3 billion, more than three times where they stood two years ago. Merged pull requests more than tripled over the same period to 1.3 billion, which works out at 3.5 million a day. On the machine side, GitHub Actions ran a job 1,440 times every second — equivalent to 124 million times a day and more than three times last year's total — while more than 30 million Dependabot pull requests were merged to keep dependencies current and reduce known vulnerabilities, about double 2025's pace.
Data source: Octoverse 2026, the annual developer-activity report published by GitHub. AIdataNewsGlobal has reorganized the published figures by theme and left every value unaltered; no additional data has been introduced in this briefing.
The mix of things being counted is itself unusual. In Nepal, one developer built a flood bulletin on GitHub Pages carrying verified information on roads, rescue, weather and power; it later grew into Gati, a Nepali-first relay of official disaster alerts running on GitHub Pages and Actions. At NASA's Jet Propulsion Laboratory, software is being made to run programs aboard spacecraft, while hobbyists produced open-source firmware letting people customize their e-readers.
Timing is the part analysts should note. Telemetry in the report shows activity climbing steadily from January, then surging around June, and the report links that bend to a cluster of industry-wide shifts arriving together: stronger underlying models, the harnesses and tools assembled around them, and the agents running on both. Agents stopped behaving like single-session tools and became persistent ones able to track progress across days and orchestrate other specialized agents, while also leaning on the cloud and executing work remotely as well as on local machines.
Two participation profiles account for the lift. The first is greenfield: from June onwards, about 55% of all commits came from brand new repositories. The second is tenured — developers who pushed code in the past 12 months average 36 months on the platform; of those who opened a pull request in August, more than 75% pushed code again in September; and the average developer now works across 10% more repositories than a year ago. Open source absorbed the same energy, with public contributions soaring to 2.5 billion in 2026 against 1.12 billion the previous year.
Named cases in the report illustrate maintenance rather than scale. @kinlane made more than 1 million contributions across his public repositories, using agentic workflows to keep a machine-readable catalog of public API information current. @kim-em logged more than 178,000 contributions while maintaining mathlib, a community library of computer-checked mathematics. @andig, a maintainer of evcc, the open source EV charging controller and home energy management system, contributed over 21,000 times. GitHub closed its account of the year by addressing exactly this long tail of maintainers, saying it sees the work being done.
Geography: builders arriving from more places
The population behind these numbers is now more than 225 million developers, and the two largest communities kept compounding. The United States stands at 38.3 million and India at 32 million, each adding more than 10 million developers this year; Brazil added 3.28 million, the third largest national gain.
| Country | Total number of developers |
|---|---|
| United States | 38.3 M |
| India | 32.0 M |
| China | 12.4 M |
| Brazil | 10.2 M |
| Japan | 7.15 M |
| United Kingdom | 6.47 M |
| Indonesia | 6.34 M |
| Germany | 6.30 M |
| Russia | 5.07 M |
| France | 4.75 M |
The forward path is what moved. Last year the report projected India reaching 57.5 million developers and the United States 54.7 million by 2030. One year into that five-year window, every country in the projections has closed at least a fifth of its own distance — 38.6% for the United States, 28.3% for India.
Regional context in the report, restated as a checklist for readers scanning cross-border dynamics:
- Americas — driver: fintech and digital payments. Brazil hosts a mature, fast-growing fintech sector; Canada shows steady AI and software investment.
- EMEA — driver: AI adoption in manufacturing and startups, plus a fast-growing developer community in Africa. Morocco contributes a fintech and startup scene; Germany enterprise AI adoption across its industrial heartland; France and Spain AI and deep-tech startups; Kenya, Nigeria and South Africa fintech, agritech and coding bootcamps.
- APAC — driver: government investment in AI infrastructure and talent. Indonesia rides a booming digital economy; Japan funds AI infrastructure; South Korea runs government-backed AI talent programs; Australia builds AI, machine learning and data skills.
Predicting the next five years is now harder than the last one. More than 40 forecasting models — time-series and regression models built on historical sign-up and usage data, each assuming a different pace of growth — were run forward to 2031. About half call for extreme growth, with developer counts on GitHub more than tripling by 2031. Just under half call for substantial growth, doubling to tripling. Only three models expect anything close to modest, under 2x growth. Forecasts remain estimates, and a spread this wide shows how much the 2026 surge widened what is plausible; the shape of the trend, a developer community growing in more places at once, is the stable part.
Sector spread: what 'everywhere' now includes
In past editions, 'everywhere' meant more developers in more countries, a baseline still visible in the top 10 ranking by pushes to public repositories, adjusted by profile location:
| Rank | Country |
|---|---|
| 1 | United States |
| 2 | India |
| 3 | China |
| 4 | Brazil |
| 5 | Germany |
| 6 | United Kingdom |
| 7 | Korea |
| 8 | France |
| 9 | Japan |
| 10 | Canada |
This edition adds occupations that were never measured at all. The method is worth auditing: the industry view of public repositories follows the North American Industry Classification System, or NAICS, the framework that sorts economic activity into sectors; a language model then read random samples of public repositories created each year between 2022 and 2026, assigning each one to the industry it serves from its name, description, topics and README. Two signals were carried through the analysis — a mention of AI inside the repository, and an agent opening a pull request against it.
Those two signals diverge. Administrative and support services, mining, health care and agriculture each show more than four in 10 repositories mentioning AI, while agent-opened pull requests there stay at less than 3 percent. Construction refuses the pattern: roughly a quarter of its repositories mention AI and about 9 percent have agent-opened pull requests, the highest share recorded for any sector, with manufacturing next in line.
- Professional, scientific, and technical services: about 5 times its own 2022 base, lifting from fourth to second place behind only information.
- Mining, quarrying, and oil and gas extraction: roughly 65 times, from a small starting base.
- Administrative and support services and construction: 17 to 18 times.
- Agriculture, forestry, fishing and hunting: up two spots in the ranking; finance and insurance and health care and social assistance up one each.
- Manufacturing: the slowest of this group, yet almost tripled.
What sits behind these shifts is domain knowledge rather than career switching. A small manufacturing team turned a photograph of a workshop's tools into a bespoke drawer organizer sized for 3D printing. In health care, a research lab lowered the barrier for scientists working through very large digital tissue scans — the material pathologists examine when studying disease. A Copenhagen university lab produced a classification pipeline for footage of chickens at play, separating frolicking from worm chasing. An energy center at the University of Waikato shipped a toolkit that helps industrial plants recover heat for reuse; Bulgaria's digital ministry published an open portal for tracing who won a public contract and for how much; and a novelist released a free tool that renders a manuscript in book form while it is being typed.
Repository-as-citation is not new in professional, scientific, and technical services, the sector that grew about 5 times and reached second place. In biochemistry, the work behind AlphaFold earned Demis Hassabis and John Jumper a share of the 2024 Nobel Prize in Chemistry. In computer science, Tensor2Tensor traveled with the paper that introduced the Transformer, the architecture behind today's large language models, Attention Is All You Need. In medicine, sars-cov-2-stability carried the code and data for one of the earliest studies of how long the virus survives in the air and on surfaces. In physics, eht-imaging was one of two open-source libraries used to reconstruct the first image of a black hole.
Workflow: agents draft, developers review
Behavior, not volume, is the sharper signal. As of September, developers start more than 40% of pull requests with an agent, up from 17% in March — 110 million agent-started pull requests in September alone, about 6x March's volume. Those requests also reach further into the tree, touching multiple files in about 87% of cases against 79% of pull requests without detected agent involvement, and they land at least as fast, as the comparison below shows.
| Measure | With detected agent involvement | Without detected agent involvement |
|---|---|---|
| Merged within two weeks, February | 83% | roughly 86% |
| Merged within two weeks, September | 90% | about 86% |
| Share changing multiple files, September | about 87% | 79% |
Adoption of the hands-off modes is where the slope is steepest. Among GitHub Copilot's monthly active users, 55% now use agentic features, up from 35% a year ago; Copilot cloud agent, where a developer assigns a task and an agent works on it independently, grew 230% year over year, against 85% for agent chat. Ambition follows capability: public repositories mentioning interoperability or reverse engineering increased more than 300% year over year, from about 30,000 created in 2025 to more than 130,000 in 2026.
Review capacity is being built out in parallel. Stacked pull requests — related work split into smaller changes that are easier to review in sequence — arrived as a preview on July 30 and had drawn more than 264,000 developers by September. AI-assisted code review and GitHub Advanced Security each grew more than 90% year over year, and platform guardrails scaled alongside: push protection, which screens for exposed secrets such as API keys and passwords, ran 5.8 billion scans this year, nearly four times 2025's pace.
Model choice: price, familiarity and orchestration
Developers match models to the work they hand to agents. In September 2026, 61% of GitHub Copilot users who selected a model themselves used more than one model across their sessions, up from 50% four months earlier. The lineup widened at the same time: paid-plan developers used 29 distinct models in May and 47 in September, and no single model carried more than 25% of requests that month. Cost and functionality are weighed against need, and many developers lean toward providers they already trust.
The measurement basis matters for anyone benchmarking vendors: requests were grouped by what models are built for — deep reasoning and planning, general purpose, and fast and lightweight — and only activity where a developer explicitly selected a model was counted. That excludes auto model selection chosen by GitHub Copilot itself, GitHub and Microsoft staff usage, background utility models, duplicate requests used for monitoring and testing, and requests made with a developer's own API key. Out of every 100 developer-selected requests, 38 went to general purpose models, 37 to deep reasoning and planning models, and 25 to fast and lightweight models, so about 25% of requests go to a fast model.
Viewed as a time series over the past three months, tracking request share alongside how many developers tried each model and how many kept coming back, the report finds three patterns:
- New releases are used heavily before most developers have tried them, with early adopters consuming a disproportionate share in the first weeks.
- When a model's share of requests falls, the share of developers still using it falls more slowly, so models stay in rotation after peaking.
- Requests spread across all three groups, and few developers stay with any one kind for long.
Some of the choosing is being outsourced: in September 2026, 63% of paying GitHub Copilot developers used auto model selection at least once. Project HydraFusion began as a research preview of model orchestration that matched models to each step of planning, implementing and checking, routing almost 60,000 sessions in the CLI and the GitHub Copilot app; that capability is now generally available as GitHub Copilot Fusion, combining model orchestration with auto model selection behind the scenes.
Languages: a possible fourth influence
Language choice has historically been driven by what developers want to build, what their teams know and what existing systems require. This year's data raises a fourth candidate: the language an agent picks when a developer does not specify one.
The 2025 order held through 2026 — TypeScript first, Python second, JavaScript third — but the pace accelerated. From September 2025 to September 2026, TypeScript added 3.62 million monthly contributors, Python 3.23 million and JavaScript 1.82 million, the largest gains among the top 10 languages, while Rust grew fastest by percentage, doubling in the past year. Extending recent trends, the report sees TypeScript and Python near 13 million monthly contributors each by September 2027, JavaScript at 7.4 million and every other language in the top 10 below about 2.2 million.
The fastest-growing communities are also where AI shows up most. The report charts each language's community growth against the share of its merged pull requests with detected AI involvement, with bubble size marking scale: the top-growth clusters sit in the about 40% to 44% band on that AI axis, the next tier spans about 21% and 34%, and read across the whole field more than one in five merged pull requests carry detected AI involvement, with 24% recorded in the mid-field band of the same chart. The report does not attribute these shares to any single cause.
FAQ: what the data means for decision-makers
Why does a developer-platform report matter to non-technical readers?
Because it measures production rather than marketing. Code pushes of 17.9 billion times in the past 12 months, 34.3 billion commits and 1.3 billion merged pull requests describe how much software an economy can actually ship, and where agents already sit inside that pipeline.
Does an agent-started pull request mean an agent owns the change?
No. The report frames the agent as the first draft and review as the control point: more than 40% of pull requests started with an agent by September, up from 17% in March, and merged pull requests averaged 3.5 million a day.
What should platform and security teams do about secrets exposure?
The reported trend is that automated checks are the only scalable option: push protection ran 5.8 billion scans this year, nearly four times 2025's pace, while more than 30 million Dependabot pull requests were merged, about double 2025's rate of dependency updates.
How much weight should the 2031 projections carry?
Little as point estimates. With more than 40 forecasting models run forward to 2031, about half returning extreme growth of more than tripling, just under half substantial growth of doubling to tripling, and only three models near modest, under 2x growth, the dispersion itself is the finding.
Which industries should a data or governance team watch first?
Those with high AI mention but low agent activity, because that is where oversight lags practice furthest: administrative and support services, mining, health care and agriculture all show more than four in 10 repositories mentioning AI while agent-opened pull requests remain at less than 3 percent. Construction, at about 9 percent, is the sector where agent involvement is already most common.
What AIdataNewsGlobal will keep watching
Three follow-ups will decide whether 2026 was a step change or a spike. First, whether the agent-started share holds as the model lineup keeps widening — 29 distinct models were in use in May and 47 by September. Second, whether review tooling scales as fast as drafting, given that Stacked pull requests drew more than 264,000 developers in roughly two months and AI-assisted code review grew more than 90% year over year. Third, whether growth in sectors measured for the first time converts into sustained maintenance, the kind that keeps mathlib or evcc current, instead of remaining a burst of new repositories.
The broader signal for global AI and data markets is that capability is diffusing outward from software into mining, construction, health care and agriculture, and that the bottleneck is moving from producing change to governing it. We will report further as new releases of the underlying dataset land.
Corrections, data questions and story tips: editorial@aidatanews.com — AI & Data News, AIdataNewsGlobal. Written by David Nakamura.