Skip to content
NewSQL, Python and Power BI training for your team
Log in

AI Agents Now Start 40% of Pull Requests on GitHub

GitHub’s Octoverse 2026 shows AI agents drafting a huge share of code. Here is what it means for Nigerian developers and data analysts.

Muhammad Jamilu Abdullahi 8 min read
Navy cover reading AI now writes the first draft, with a donut chart showing 40% of pull requests started by AI agents, up from 17% in March

On 28 October, at GitHub Universe, Microsoft executive Jay Parikh showed a chart that summed up the year. For years the lines barely moved. Then, on the right-hand side, they shot almost straight up.

The number he picked out was merged pull requests: about 220 million a month, roughly three times the level of less than a year earlier.

The same day, GitHub published Octoverse 2026, its yearly report on activity across the platform. Its biggest claim is that by September, more than 40% of pull requests on GitHub started with "an agent's first draft". In March, the figure was 17%.

Put simply, AI tools that write code on their own now do much of the first drafting on the world's largest code platform. For anyone learning Python, SQL or Power BI in Nigeria, that changes what employers will expect.

What the report found

GitHub's year ran from 1 October 2025 to 30 September 2026, and it now counts private projects as well as public ones. Developers pushed code 17.9 billion times, more than in 2024 and 2025 combined. They made 34.3 billion commits and merged 1.3 billion pull requests, about 3.5 million a day.

Growth sped up sharply around June. GitHub links this to better AI models, new tools, and agents that coordinate other agents. In September alone, agents started about 110 million pull requests, around six times March's volume.

Measure2026Compared with
Pull requests started by an agentOver 40% (September)17% in March
Merged pull requests per monthAbout 220 millionAbout 3x in under a year
Code pushes, year to 30 September17.9bn6.7bn a year earlier
Commits34.3bn12.9bn
Merged pull requests1.3bn538m
GitHub Actions jobs (automated build and test tasks)45.4bn14.0bn
Developers on GitHubOver 225mOver 180m (2025 report)

Earlier-year figures are GitHub's recalculated numbers, which include private projects. The monthly figure comes from the Universe keynote.

Agents are not just making tiny tweaks. About 87% of agent-started pull requests changed more than one file, against 79% of the rest. By September, 90% of agent pull requests were merged within two weeks, compared with about 86% of others.

Among Copilot's monthly users, 55% now use its agent features, up from 35% a year ago. Agents also increasingly work in the cloud, not only on a developer's laptop.

The language rankings matter for data people. TypeScript stayed first, while Python came second and added 3.23 million monthly contributors in a year. GitHub forecasts both will have close to 13 million monthly contributors by September 2027.

A newsroom way to understand it

Picture a newspaper desk in Kano or Lagos.

A commit is a reporter saving a version of a story with a short note, such as "added the commissioner's comments". A push is sending those saved drafts from a laptop to the newsroom's shared system, which here is GitHub.

A pull request is filing the story to the editor's desk: please check this and put it in tomorrow's paper. The editor reads it, asks for changes, and only then "merges" it into the edition, meaning the live software.

A coding agent is a tireless stringer. Give it an assignment, such as "fix the broken date filter on the sales report", and it opens the files, writes the code, runs checks and files its own draft. Four in ten drafts reaching the desk now come from stringers like this. The editor still decides what gets printed.

Why it matters

The hard part of software is moving from writing code to judging it. When drafts are cheap, the scarce skill is knowing whether a draft is correct, safe and what the business actually asked for.

Even checking is being automated. In March, GitHub said its Copilot code review tool had passed 60 million reviews and was doing more than 20% of code reviews on the platform.

Agents also lower the barrier to starting. From June, about 55% of commits went into brand-new repositories.

For analysts, Python's rise is good news. The language behind pandas, notebooks and machine learning is now one of GitHub's two biggest, and agents can take on dull jobs like cleaning data and wiring up pipelines.

The other side

These are GitHub's own figures. GitHub is owned by Microsoft, which sells Copilot, and no outside body audits them. GitHub also changed its method this year: its 2025 report gave 518.7 million merged pull requests, while the 2026 report restates 2025 as 538 million.

The 40% figure needs care too. It only counts agents GitHub can detect, so GitHub calls it a floor. Yet starting a draft is not the same as writing all of it, or writing it well.

Reviewers are feeling the strain. In February, The Register reported open-source maintainers complaining about a flood of weak AI-assisted contributions; one called a documentation change "plausible nonsense". Within a fortnight, GitHub let projects switch off pull requests or limit them to trusted collaborators.

Independent research is mixed. Researchers at Drexel University and Missouri University of Science and Technology studied 33,596 agent-written pull requests in popular projects and found about 71% were merged. Rates ranged from 43% to 83% by agent, and rejected ones were often left without real reviewer attention, duplicated other work or failed automated tests.

The platform itself has creaked. On 17 August, GitHub suffered an outage of 7 hours and 47 minutes after traffic hit a new peak. Its chief technology officer, Vlad Fedorov, wrote: "We failed to scale critical components before demand exceeded their capacity."

AI is speeding up attackers as well. In late October, serious flaws in PHP, which runs a large share of the world's websites, were patched after being found with AI help, and exploit attempts followed within hours. The scramble has been nicknamed Black October.

Then there are jobs. A 2025 Stanford study of US payroll data found workers aged 22 to 25 in the most AI-exposed occupations, including software development, saw a 13% relative fall in employment. That is American data, but Nigerian juniors increasingly compete in the same global market.

What it means for Nigeria and Africa

Octoverse 2026 calls Africa a fast-growing developer community and names Kenya, Nigeria and South Africa for their fintech, agritech and coding bootcamps. But it gives no developer count for any African country this year.

GitHub's latest published Nigeria figure came in 2024: more than 1.1 million developers, up 28% in a year and the largest of the African countries it listed. Egypt had more than 990,000 and South Africa more than 664,000. In 2025, GitHub named Nigeria a standout market in Africa and the Middle East, a region that added 3.4 million developers.

India shows what scale looks like. It now has 32 million developers on GitHub and added more than 10 million this year.

At home, the federal 3MTT (Three Million Technical Talent) programme had trained more than 135,000 Nigerians by January, with 1.8 million in its pipeline, ITWeb Africa reported. Its courses include data analysis and visualisation. The minister, Bosun Tijani, said the next phase would focus on "deepening impact rather than just expanding numbers". The Mastercard Foundation's Scaling ALX programme, which runs in Nigeria, aims to give tech skills to 2.4 million young Africans by 2030.

Cost matters too. Copilot Free offers 2,000 code completions a month with limited agent access, and verified students get a free plan. Copilot Pro costs $10 a month, about ₦13,300 at the official rate in late October, and Pro+ costs $39. Running Copilot's agents in the cloud needs a paid plan.

The opening is real but narrow. Agents make it cheaper for small Nigerian teams to build products, but they also make routine coding cheaper everywhere.

What you can do now

  1. Learn to review code. Read other people's pull requests on GitHub and ask: does this do what it claims, and what could break? Teams now need this skill most.
  2. Write tests and checks. Tests are how you prove an agent's work is right. In data work, that means checking row counts, missing values and totals before a report goes out.
  3. Get your Python and SQL basics solid. You cannot direct or correct an agent in a language you do not understand.
  4. Master Git and GitHub. Practise commits, branches and pull requests on your own projects. A tidy profile with clear READMEs is a portfolio employers can check.
  5. Practise directing agents. Start on a free tier, give small, clear tasks, and review every result line by line. Note which instructions work best.
  6. Contribute to open source with care. Documentation fixes merged most often in the Drexel study, so they are a sensible first step. Follow each project's rules, say when you used AI, and never submit work you have not checked.
  7. Watch the independent data. Look out for outside studies of agent quality, GitHub's reliability, dollar price changes, and whether GitHub publishes Nigeria figures again.

Agents can now file the first draft. The people who do well will be the ones who know how to edit it.

Sources

Was this helpful?
#GitHub#Octoverse 2026#AI agents#Software development#Python#Careers in tech
Learn it live, with an instructor

SQL and Python for Data Analysis

A live programme. Each learner builds their own MySQL database, fills it with real data, then analyses it in Python.

  • Live cohort · Starts 4 Nov 2026
  • ₦200,000, or 4 × ₦50,000
View programme
Get new posts by email

When we publish a new guide, we'll send it to you. No spam, and you can unsubscribe at any time.

We'll email you a link to confirm. See our Privacy Policy.

Keep reading

All posts