Anthropic's code is now written so fast that git broke, so seven engineers and Claude built a replacement in five months

Anthropic said on Tuesday, Oct. 13, 2026, that it has moved its main code repository off git onto Scion, an in-house version control system built for AI agents rather than people. A team of seven built Scion, a code-review tool called Antique and half a dozen related systems in five months, largely with Claude, while the volume of code Anthropic merges each day grew tenfold.

Git has been the plumbing of almost all software for twenty years, built on the assumption that a person sits at a laptop, sometimes offline, writing code by hand. Anthropic just told us that assumption broke inside one company in about four months. Twelve times more code a day is not a productivity stat, it's a different kind of workplace, and the tooling built for people is the first thing to crack. Two caveats: this is Anthropic describing its own success with its own product, and it is unusually extreme, as the company admits. But the most striking line isn't about git at all. It's that every engineer on the team now rates the AI as a better coder than themselves, and they still shipped in five months what used to take years, by deciding what to build. That's what the job is turning into.

Anthropic averaged around 2,500 code changes merged per day in May 2026 and more than 30,000 per day as of September. It says source control became its biggest development bottleneck: “coding is no longer the bottleneck,” because Claude can autonomously iterate on research loops and build services, so the limit is developer infrastructure built for human scale — source control, builds and CI. Source control is the system that stores every change to a codebase. CI, continuous integration, is the automated tests that run on every change. Those lines are Anthropic’s.

Its monorepo’s main-branch history is almost 100 GB and a compressed working copy over 25 GB. A monorepo is one giant repository for all of a company’s code. It runs on the order of 25 million code checkouts a day for CI alone, which fetched naively would be more than 600 petabytes of git traffic a day. A petabyte is a million gigabytes. At peak it ran several hundred internal git mirrors. Git is the version-control tool almost every software team uses. Those lines are Anthropic’s.

Scion is designed on the assumption that nearly all clients are agents and jobs that are always online: the server answers history and blame questions, creates commits and performs merges and rebases, while the client keeps a bounded local cache. Blame is the history of who changed each line. A merge or rebase is a way of combining branches of work. All source control queries are exposed by API, so many jobs need no local repository. Anthropic cites Google’s Piper and Meta’s Sapling as prior art — in-house systems those companies built when git stopped fitting. Those lines are Anthropic’s.

Results, per Anthropic: Scion’s infrastructure footprint is 200 times smaller than the previous git mirroring setup in CPU and memory; CI and research jobs check out code in seconds rather than minutes; CI gained 25% CPU capacity; each cache tier holds a hit rate over 99%. Those figures are Anthropic’s.

A team of seven built Scion and Antique, a code-review tool, plus code search, a source control permissions system and other tools, and fully migrated the monorepo from git in five months, while daily merge volume rose tenfold. Anthropic says this would previously have taken “years of iterative migration, with many more engineers.” Those lines are Anthropic’s.

Anthropic said: “All of the engineers who worked on this system would now rate Claude as better at writing code than them. But our engineers still supplied the invaluable understanding of the high-level inter-system components and the design-level tradeoffs.” In one example it describes, Claude rewrote a component with no further instructions into a version that used 20 times less memory and ran 4 times faster. Those lines are Anthropic’s.

Anthropic says it started building Scion around the time Claude Mythos Preview and Claude Tag became available internally, and that it expects other companies using AI coding tools to hit the same pressure points before long. Those lines are Anthropic’s.

In plain terms, Anthropic said on Tuesday that git could not keep up after daily code merges jumped from about 2,500 to more than 30,000, so seven engineers and Claude built Scion in five months and moved the whole company onto it. The new system is 200 times smaller to run, Anthropic says, and every engineer on the team now rates Claude as a better coder than themselves.