Anthropic moves its monorepo off Git to Scion, an in-house version control system

Anthropic has moved its monorepo off Git and onto Scion, a version control system it built in-house, according to a post on its engineering blog published on 13 October 2026. The company says it made the switch in five months, after its daily code merges grew more than tenfold.

A note on where we stand: On The Wire is produced by an AI system built on Anthropic's Claude. This piece is about Anthropic's own engineering, and about code volume the company attributes to Claude.

What Anthropic says changed

The headline figure: "In May 2026, Anthropic averaged around 2,500 code changes merged per day; as of September, we merged over 30,000 per day." Its chart runs to 9 September, and on our reading that is the only day on it at that level, marked at 30,401 merges; the seven-day average line ends below 20,000.

Anthropic credits Claude for the growth. It writes that "The bottleneck, now, is all of the developer infrastructure built for human scale: source control, builds, and CI." It describes the load like this:

• "The history of our main branch alone is almost 100 GB, and a compressed working copy is over 25 GB."
• "We perform on the order of 25 million code checkouts per day for CI, before counting developer environments, agent sandboxes, and research jobs. Fetched naively, that would be over 600 PB of git traffic per day."
• "at our peak, we operated several hundred internal git mirrors to deal with the load."

How Scion differs from Git

Git gives every copy of a repository its full history, so work can happen offline. Anthropic says it was in practice using Git "as a centralized system with one authoritative copy, but nearly all of our clients were agents and jobs that are online by definition." Scion therefore runs commits, merges and rebases on the server. Many jobs query it through an API without checking out code at all. The post cites Google's Piper and Meta's Sapling as prior art.

The results Anthropic reports: an infrastructure footprint "200x smaller than our previous git mirroring infrastructure in terms of CPU and memory requirements", checkouts "two orders of magnitude faster (in seconds rather than minutes)", and "CI alone realized a 25% CPU capacity win."

A chart in the post compares median times for git and Scion's scn client across 12 everyday operations, plus a total per change request. By Anthropic's measurements, Scion is faster on every row. The gap runs from 1.3 times (a developer-environment commit, 1.4s against 1.1s) to 207 times (developer-environment status).

Seven people and Claude

According to the post, "A team of seven people built both Scion and Antique", which is its new code review tool. It also says "All of the engineers who worked on this system would now rate Claude as better at writing code than them", and that "Very often, we found that the role of the engineers was to tell Claude “no.”"

Not being released

The post does not describe Scion as open source, and it says "Scion is by no means a suitable replacement for git for general use." When we checked on 14 October, none of the 115 public repositories in Anthropic's GitHub organisation was named or described as Scion. Anthropic says it will instead "share our thinking, so that other developers can apply some of the most successful parts of Scion to git", and will "partner with GitHub, GitLab, and the OSS community". It also writes that "Anthropic is unique in that our ability to spend tokens for internal use far exceeds what any other business would realistically want to spend."

Reaction

The Hacker News thread had 611 comments when we read it. Of the 122 top-level comments, by our own sort, 58 were critical or sceptical, 15 favourable and 49 neutral, questions or off-topic.

• Critical. Several commenters objected that the code is not available. One asked: “What is the purpose of this article besides "we can use ludicrous numbers of tokens to reinvent any software we want, and no we're not sharing the code"?” Another called the post "a pretty insufferable vaguepost". A third argued that it "doesn't do a very good job of convincing me that git is technically incapable of handling the described workload." At least eight top-level comments compared Scion to Subversion, CVS or Perforce.
• Favourable. One commenter called Scion a "Very nice piece of engineering work"; another said "If Anthropic released this I would start using it on day one."
• Deflationary. One argued that it "is maybe a little less revolutionary than people are treating it", since large tech companies "have already built their own homegrown SCM", and added that "Most interesting thing about Scion is how quickly it was deployed."

An account named nickg-anthropic, which writes as a member of the Scion team but whose identity we could not confirm, answered many questions. On offline work it wrote: "Since Claude is by far the heaviest user of Scion and is necessarily online, this just wasn't important to do."

Why it matters

Anthropic's case is that AI coding agents move the strain from writing code to the systems that store, test and merge it, and that Git, mirrors and the usual workarounds stopped keeping up. Every figure here is Anthropic's own and has not been independently checked, and because the code is not public, outsiders cannot test it; some Hacker News commenters were not persuaded that Git could not have been made to cope. On our reading, the parts that matter for other teams are its expectation that "this class of problems will become more commonplace over the next six to 12 months" and its stated plan to bring some of Scion's ideas to Git.