GitHub Says More Than 40% of Pull Requests Now Start With an Agent. TypeScript Is the Language They Gravitate To When You Don't Choose.
GitHub's Octoverse report, published October 28, says that as of September developers start more than 40% of pull requests with an agent. In March 2026 that share was 17%. In absolute terms, agent-started pull requests reached 110 million in September, roughly six times March's volume.
I read a lot of year-in-review reports and most of them are vanity metrics. This one has a number that changes a decision I make on every new project: which language to start in.
The headline numbers
For the twelve months ending September 30, GitHub counts 17.9 billion pushes, 34.3 billion commits, and 1.3 billion merged pull requests, about 3.5 million per day. It says it now serves more than 225 million developers, with the United States at 38.3 million and India at 32.0 million. Public contributions hit 2.5 billion in 2026, against 1.12 billion the year before.
On the agent side, GitHub reports that agent-started pull requests change multiple files more often, about 87% of the time versus 79% for pull requests without detected agent involvement. Among Copilot's monthly active users, 55% now use agentic features, up from 35% a year earlier. And 61% of Copilot users who chose a model used more than one in September.
One more figure I find useful: among repositories created this year, those with an agent configuration file logged eight times as many commits as those without one. That is a correlation, and repos that bother to write a config file are probably more active to begin with. But it fits what I see when I give an agent written instructions instead of a blank prompt.
The language signal
The line I'd underline is about TypeScript. GitHub ranks TypeScript first among languages by contributors, ahead of Python and JavaScript, and it grew by 3.62 million monthly contributors between September 2025 and September 2026. More interesting, TypeScript was the primary language in 35% of 2026 repositories with an observed agent signal, versus 24% of those without one.
The report itself suggests agents may favor a narrower set of languages when developers leave the choice open. GitHub does not claim causation, and I'd be careful to say the same. It could be that developers who use agents also prefer TypeScript for its type checks, or that agent tooling ships with TypeScript templates. What the data does show is that the two travel together.
What it means for a solo stack
When I start a project now, the first draft often comes from an agent. That makes the language a bet on two things at once: how well the agent writes it, and how well the compiler catches what the agent got wrong.
TypeScript scores on both. A type error is a cheap, automatic review comment, and it fires before the code reaches a human. For someone with no teammate to review a 15-file agent diff, that matters more than syntax preferences. The 87% multi-file figure is the reason: a diff that touches many files is exactly where a type checker earns its keep.
Rust doubled in the past year, the fastest percentage growth among top languages in the report, and the same logic applies there for a different reason, because the compiler is strict. I would not rewrite anything because of that. I would use it as a tiebreaker for new, small tools.
Where this could mislead you
Don't read the TypeScript number as "agents are bad at everything else." Python is second by contributors and gained 3.23 million monthly contributors, and for data work it is still the obvious choice. The 35% versus 24% gap is about default choices in open-ended situations, not about capability.
Also, 40% of pull requests starting with an agent says nothing about quality. Started is not merged, and the numbers I saw don't say how many of those 110 million pull requests were accepted, so I wouldn't assume the merge rate matches.
What I'd actually do
Three things. First, if a project has no hard language constraint, I'd default to TypeScript with strict mode on and let the compiler act as the first reviewer. Second, I'd add an agent configuration file to every repo on day one, with the stack, the test command, and the conventions, because the eight-times commit figure points that way even if the causation is loose. Third, I'd keep a habit of reading the diff stat before the code: if an agent-started change touches many files, I run the type check and tests before I read a line.
The share of work that begins with an agent is only going up. The useful response is to make the cheap automatic checks do more of the reviewing.
Author
Lukas
@lukcombinator