Anthropic Scion is the company’s answer to a problem no engineering team has ever had before: its AI agents write so much code that git itself broke. On October 13, 2026, Anthropic revealed on its engineering blog that it had fully migrated its monorepo from git to an in-house version control system called Scion — and the numbers behind the switch are staggering.
In May 2026, Anthropic merged around 2,500 code changes per day. By September, that figure had passed 30,000 per day — a more than twelvefold jump in four months, driven by Claude-powered coding agents. Writing code stopped being the bottleneck. The plumbing around the code became the bottleneck instead.
“The bottleneck, now, is all of the developer infrastructure built for human scale: source control, builds, and CI,” Anthropic wrote in its engineering post “Redesigning source control at Anthropic”. For anyone building a business on AI coding tools, that sentence is the whole story.
The numbers that broke git
Anthropic’s scale figures explain why incremental fixes ran out. The history of its main branch alone is almost 100 GB, and a compressed working copy is over 25 GB. The company performs on the order of 25 million code checkouts per day for continuous integration — before counting developer environments, agent sandboxes, and research jobs. Fetched naively, Anthropic says, that would be over 600 petabytes of git traffic per day.
At its peak, the company ran several hundred internal git mirrors to carry that load. But git’s core design assumption — every clone is a complete copy that can work offline — made it the standard for humans and the wrong tool for agents. Most of Anthropic’s clients are agents and automated jobs that are online by definition and need one diff, not the full history of the company.
What Scion does differently
Scion flips git’s assumptions. It presumes clients are always online and moves the work to the server: the server answers questions about file history and blame, and performs commits, merges, and rebases. Clients keep a bounded cache and fetch only the data they need. Checkouts are sparse and can be virtualized, and many tools can query or change code entirely through an API.
The results, per Anthropic’s post: Scion’s infrastructure uses 200x less CPU and memory than the old git-mirroring setup; CI and research checkouts are two orders of magnitude faster (seconds instead of minutes), with each cache tier above 99%; and CI alone gained 25% in CPU capacity.
Alongside Scion, a team of just seven people also shipped a code-review surface called Antique, plus code search and new tooling for landing large volumes of reviewed code — all in five months, while merge volume was still climbing. In the past, Anthropic notes, a project like this would have taken a company years with many more engineers.
The twist: Claude built the thing that replaced git
Claude’s ability to write code was both what created the bottleneck and how the team escaped it. 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 ran 4x faster — with no further instructions.
The human role changed in the process. Anthropic says all of the engineers who built Scion would now rate Claude as a better coder than themselves, and that their job very often became telling Claude “no.” Team members still review every change request, but they examine the intent of the change rather than reading every line.
The money angle: where the value moved
The Scion story marks a genuine economic inflection point. The scarce resource in software is no longer keystrokes — it is review, infrastructure, and judgment.
1. The bottleneck moved downstream. When code generation gets cheap, the expensive parts become everything around it: source control, CI, testing, and review. Teams budgeting for AI coding tools should budget at least as much for the pipeline that absorbs the output.
2. Review is the new coding. Anthropic’s engineers shifted from writing code to reviewing intent and rejecting bad changes. If you sell developer tools — or manage a dev team — the highest-leverage investment right now is review automation: better diffs, adversarial review loops, and post-merge testing. Anthropic says a traditional merge queue “doesn’t make much sense” at agent scale and invested in testing after changes merge instead.
3. Dev-tool platforms are on notice. Anthropic says Scion is “by no means a suitable replacement for git for general use,” but it wants to partner with GitHub, GitLab, and the open-source community on server-side RPCs for agent-scale workloads. The company expects this class of problems to become commonplace within six to 12 months. If your business runs on git-based CI, watch this space.
4. Small teams can build like giants. Seven people rebuilt source control in five months because Claude did much of the heavy lifting. That same leverage is available to any small team right now.
Frequently asked questions
What is Anthropic Scion?
Scion is Anthropic’s in-house version control system, built to replace git for its own monorepo. It is server-centric and API-first: a central service handles commits, merges, and history while clients fetch only what they need.
Is Scion replacing GitHub?
No. Anthropic calls it an internal system, not a product, and says it is “by no means a suitable replacement for git for general use.” The company wants to share its approach with GitHub, GitLab, and the open-source community instead.
Why did git fail at Anthropic?
Git assumes every clone is a complete, offline-capable copy. With ~25 million checkouts per day and most clients being always-online agents that need only a fraction of the code, that design became hundreds of petabytes of daily traffic.
What should developers do about this?
Measure your full code-to-deploy loop, not just typing speed. Invest in automated review and post-merge testing. And expect source-control and CI costs to grow as agents write more of your code — Anthropic says most companies will feel this pressure within a year.
Keep Reading
Source: Anthropic Engineering, “Redesigning source control at Anthropic” (October 13, 2026). All figures are Anthropic’s own.