Claude Writes So Much Code That Anthropic Had to Replace Git

Anthropic has replaced git, the tool most of the world’s developers rely on, with its own system called Scion, because its AI agents now write more code than git can handle.

The company shared the details on October 13, 2026, on its engineering blog. Writing code is no longer the bottleneck, it says; the problem is developer infrastructure built for human scale, including source control, builds and continuous integration (CI).

From 2,500 to 30,000 merges a day

In May 2026, Anthropic merged about 2,500 code changes per day on average. By September it was merging more than 30,000, and the company does not expect that growth to slow.

Metric	According to Anthropic
History of the main branch	Almost 100 GB
Compressed working copy	Over 25 GB
Code checkouts per day for CI	On the order of 25 million
Daily traffic if every checkout were a full fetch	Over 600 PB

At its peak, the company ran several hundred internal git mirrors to keep up. That bought time, but it stopped scaling as the merge rate kept climbing.

How Anthropic Scion works

Git is designed so every developer holds a full copy of the repository and can work offline. At Anthropic, nearly all of the clients are agents and automated jobs that are always online.

So in Scion, the server does most of the work: it creates commits, performs merges and answers questions like who last changed a line. A local machine only fetches the files it actually needs.

Anthropic says it built on lessons from similar systems at other tech giants, notably Google’s Piper and Meta’s Sapling.

The payoff: 200x less infrastructure

According to the company, Scion’s footprint is 200x smaller than its old git mirroring setup in CPU and memory. CI and research jobs now check out code in seconds instead of minutes, and CI alone freed up 25% of its CPU capacity.

Alongside Scion, the team built a code review tool called Antique, code search and a detailed permissions system. All of this happened while daily merges grew tenfold.

What Claude did, and what was left to humans

One example from the post: the team asked an early internal snapshot of Claude Mythos Preview to rewrite a component that had been written by hand. The new version used 20x less memory and ran 4x faster.

Every engineer on the project now rates Claude as better than them at writing code, Anthropic says. Their job has shifted to design decisions and, quite often, telling Claude “no.”

Claude helps with operations too. Copies of Claude in Slack channels each watch a single component of the system, and Anthropic calls them “superfans.”