Anthropic’s Scion Explained: Replacing Git at 30,000 Merges

Anthropic says it merged more than 30,000 code changes a day into its monorepo in September 2026, up from around 2,500 a day in May, and that the jump led it to replace git with an in-house version control system called Scion. It described the system in an Engineering blog post on 13 October 2026.

In the post, Anthropic says writing code is no longer what holds it back: “The bottleneck, now, is all of the developer infrastructure built for human scale: source control, builds, and CI.”

What Anthropic Built

Anthropic says a team of seven people built Scion and fully migrated its monorepo off git in five months. In the same period, the post says, the team also built a code review tool called Antique, code search and a source control access-control system.

The post says the team did this while the volume of code merged per day increased by 10x. Anthropic says that in the past, developing and deploying a source control system like this would have taken a company years, with many more engineers.

What Was Posted on X

Anthropic’s own announcement of the Engineering post, posted on 13 October 2026. It says the company merged around 2,500 code changes a day in May and more than 30,000 a day in September, that it became bottlenecked by developer infrastructure built for human rather than agent scale, and that it built a new version control system and fully migrated its monorepo from git in five months.

Why Git Stopped Fitting

Anthropic says the history of its main branch alone is almost 100 GB, and a compressed working copy is over 25 GB. It says it performs on the order of 25 million code checkouts per day for CI, which, fetched naively, would be over 600 PB of git traffic per day.

At its peak, the post says, Anthropic operated several hundred internal git mirrors to deal with the load. It says that nearly all of its git clients were agents and jobs that are online by definition, so the features that make git good for offline, distributed work were working against it.

How Scion Works

According to the post, Scion assumes clients are always online. The server answers questions about file history and blame, creates commits and performs merges and rebases, while the client keeps a bounded local cache and fetches data when it needs it.

Anthropic says the unit of work is a “changeset”, an entry in a write-ahead log, and that changing a ref is a compare-and-swap on a single Spanner row. Checkouts are sparse by design and can be virtualized, so a job only fetches the files it reads, the post says.

What It Changed

Anthropic says its Scion infrastructure footprint is now 200x smaller than its previous git mirroring infrastructure in terms of CPU and memory requirements. CI and research jobs check out code two orders of magnitude faster, in seconds rather than minutes, and CI alone realized a 25% CPU capacity win, according to the post.

The post says each tier of the checkout system keeps a cache hit rate of over 99%.

How Claude Was Used

Anthropic says that when it asked an early internal snapshot of Claude Mythos Preview to rewrite its handwritten git-to-Scion ingestion pipeline, the model produced a version that consumed 20x less memory and was 4x faster.

The post says all of the engineers who worked on the system would now rate Claude as better at writing code than them, and that very often the engineers’ role was to tell Claude “no.” Team members still review every change request, but they now examine the intent of the change as opposed to reading every line of code, according to Anthropic.

Not a Product

Anthropic says Scion is by no means a suitable replacement for git for general use, and that it is difficult to offer Scion in a form that would be broadly useful to other organizations.

The company says it anticipates this class of problems will become more commonplace over the next six to 12 months, and that it wants to partner with GitHub, GitLab and the open-source community, sharing its thinking so other developers can apply parts of Scion to git.

GitLab’s Answer: Keep Git, Rebuild Underneath

GitLab answered in a 13 October blog post, “Scaling Git for the agentic software factory”. What follows is GitLab’s own account, and it doubles as a pitch for its product. GitLab says Scion is purpose-built for enormous monorepos where nearly all clients are agents that are always online, and that Anthropic acknowledges its approach isn’t a general-purpose replacement for Git and requires its own code review, search and CI infrastructure.

GitLab’s alternative, which it calls next-gen SCM, keeps the Git protocol and changes the architecture underneath: stateless compute that scales horizontally against shared object storage, and agent-native interfaces that let agents request the files and history they need without repeatedly cloning entire repositories. GitLab says a single agent invocation on a traditional Git workflow can involve 5–10 GB of data transfer and more than 30 seconds of setup.

In early internal tests, GitLab says, workloads completed up to 50x faster, used up to 2x fewer tokens and generated up to 1,000x less network traffic than conventional workflows. GitLab’s own caveat: “These are peak improvements under specific test conditions, and we’re continuing to validate performance under sustained, concurrent load.” Next-gen SCM is in private beta.

GitLab also says that over the past year its customers created 40% more CI/CD pipelines on GitLab.com, while code pushes increased 50%. It announced next-gen SCM in June at GitLab Transcend, joined on stage by Nick G. from Anthropic’s source control team, and says the two have continued collaborating on agent-scale development. GitLab quotes him: “Scaling our monorepo for agentic development has changed how we think about Git server architecture. We see room to expand what Git servers can do for these workloads without departing from the overall Git model.”

As we read it, the split is between a custom system built for one company’s giant, always-on monorepo and a Git-compatible rebuild aimed at enterprises that can’t move source control onto new infrastructure. GitLab’s numbers are its own and have not been independently tested.

The Structural Read

Anthropic says that nearly all of its git clients were agents and jobs that are online by definition, and that git’s design for offline, distributed work was undermining its ability to scale.

The post says Scion moves work to the server: history, blame, commits, merges and rebases are answered or performed there, and jobs can often skip a local checkout entirely by calling the API.

As we read it, the post also describes a change in the human job: Anthropic says its engineers now review the intent of each change rather than every line, and that their role was very often to tell Claude “no.”

Three Implications

CODE HOSTS Anthropic says it wants to partner with GitHub, GitLab and the open-source community, and that git platforms can provide comprehensive server-side RPCs like Scion.

CI AND REVIEW The post says a proper merge queue does not make much sense at agent scale, and that Anthropic has instead invested in post-merge and pre-deploy automated testing.

ENGINEERING TEAMS Anthropic says a team of seven built Scion and Antique and migrated in five months, which it says would in the past have taken a company years with many more engineers.

RELATED On agents working at volume: Factory and Cognition ship rival tools to run agent fleets and XBOW wins Google’s $250,000 Chrome bounty with GPT-5.6 Sol. More Anthropic news from this week: Anthropic turns away revenue to cut distillation up to 99%; Anthropic says self-replicating AI could run on $55 TV boxes; and Anthropic to share first Q3 financials with IPO investors.

The Business Engineer Lens

This story maps onto the Business Engineer framework The Agentic Harness War.

The framework puts it this way: “The surface, not the model, governs autonomy.”

As we read it, Scion is that argument applied to the plumbing: Anthropic says the agents were already writing the code, and what had to be rebuilt was the source control, review and CI they run through.

What Is Not Established

All the figures in this piece are Anthropic’s own and have not been independently verified. The post does not say what share of the 30,000 daily merges were written by Claude rather than by people, and it gives no cost figures for building or running Scion.

We read Anthropic’s Engineering post and its X announcement in full on 13 October 2026.

The Bottom Line

Anthropic says its daily code merges rose from around 2,500 in May to more than 30,000 in September, and that a team of seven replaced git with Scion in five months to keep up. The company says Scion is internal, not a product, and that it expects other companies using AI coding tools to hit similar limits within six to 12 months.