# Anthropic Scion: Why It Replaced Git at 30,000 Merges a Day

> Anthropic built Scion, its own source control, after daily merges rose from 2,500 to 30,000 in four months. What changed and who it is for.

- Author: Nitish Garg, Founder & CEO, CellCog
- Published: 2026-10-13
- Canonical (HTML): https://cellcog.ai/blog/anthropic-scion/
- Section: Guides / Choosing a platform
- Publisher: CellCog (https://cellcog.ai), the AI employee platform. Blog index for agents: https://cellcog.ai/blog/llms.txt

## Key points

- Anthropic merged about 2,500 code changes a day in May 2026 and more than 30,000 a day by September, most of them written with Claude.
- It replaced Git in its monorepo with Scion, an in-house system built on the idea that agents and jobs are always online, so the server does the heavy work and clients fetch only what they need.
- A team of seven built Scion and a code review tool called Antique and finished the migration in five months.
- Anthropic says its engineers now rate Claude as a better coder than themselves, and that much of their job was telling Claude no when it broke system constraints.
- Scion is not offered to anyone else. Anthropic says it is not a general Git replacement and expects other teams to hit the same limits within six to 12 months.

## At a glance

- **What is Anthropic Scion?** An in-house version control system Anthropic built to replace Git in its monorepo, described in an engineering post on October 13, 2026.
- **Can I use Scion?** No. Anthropic has not released it and says it is not a suitable Git replacement for general use.
- **Why did Anthropic build it?** Merges grew from about 2,500 a day in May 2026 to over 30,000 in September, and Git mirrors stopped scaling for agents that check out code millions of times a day.

**Anthropic wrote so much code with Claude that Git stopped keeping up, so it built its own source control.** In an [engineering post on October 13, 2026](https://www.anthropic.com/engineering/redesigning-source-control), the company said it now merges more than 30,000 code changes a day, up from about 2,500 in May, and that "coding is no longer the bottleneck to building systems or doing research." The new bottleneck was the infrastructure around the code. Its answer is Scion, an in-house version control system that replaced Git in its monorepo in five months.

Scion is not a product and is not for sale. The post is a case study in what happens to developer tooling when most of the people writing code are agents.

## Why Git stopped scaling for Anthropic

Git gives every clone a full copy of the repository so it can work offline. Anthropic's clients are almost all agents and jobs, which are always online, and its repository outgrew that model.

*Table: Anthropic's monorepo and Scion, by the numbers (from Anthropic's post)*

| Measure | Figure |
|---|---|
| Merges per day, May 2026 | About 2,500 |
| Merges per day, September 2026 | Over 30,000 |
| Main branch history | Almost 100 GB |
| Compressed working copy | Over 25 GB |
| Code checkouts per day for CI | About 25 million |
| Git traffic per day if fetched naively | Over 600 PB |
| Git mirrors at peak | Several hundred |
| Migration time | Five months, team of seven |

Anthropic says it had already used the usual workarounds, partial clones like Git's [partial-clone feature](https://git-scm.com/docs/partial-clone), mirrors and repository snapshots, and that they bought time but grew harder to run as merges climbed. It also says it did not expect the growth to slow: "the latent demand for merging code is practically infinite."

## How Scion works

Scion borrows from earlier systems for huge monorepos, naming [Google's Piper](https://cacm.acm.org/research/why-google-stores-billions-of-lines-of-code-in-a-single-repository/) and [Meta's Sapling](https://sapling-scm.com/), and turns Git's design around:

*Table: Git and Scion, as Anthropic describes them*

| | Git | Scion |
|---|---|---|
| Client | Full copy, works offline | Always online, fetches what it needs |
| Who does the work | Every clone can compute history and merges | The server answers history, blame, commits, merges and rebases |
| Commits | Made locally, then pushed | Made on the server, nothing to push |
| Checkouts | Full or partial clones | Sparse or virtual by design; many jobs need none |
| Scaling | Mirrors and packfiles | Distributed, content-addressed data and tiered caches |

The results Anthropic reports: an infrastructure footprint 200x smaller than its old Git mirrors in CPU and memory, cache hit rates over 99% at every tier, checkouts in seconds rather than minutes, and a 25% CPU capacity win for CI.

## What Anthropic says about engineers

The most quoted lines are about people. "A team of seven people built both Scion and Antique," Anthropic writes, Antique being its new code review tool, while daily merges grew tenfold. It says "All of the engineers who worked on this system would now rate Claude as better at writing code than them," and that their value moved to system design and tradeoffs, often by stopping Claude from breaking constraints or over-engineering in response to hypothetical concerns.

One example from the post: an early internal snapshot of Claude Mythos Preview rewrote a handwritten ingestion pipeline with no further instructions and produced a version that used 20x less memory and ran 4x faster. Anthropic also runs Claude instances it calls "superfans," each assigned to watch the telemetry of a single component.

## Who it is for

Anthropic is direct that this is not a general tool: "Scion is by no means a suitable replacement for git for general use." It says the ideas only make sense for enormous monorepos, and that its token spending on internal work is far beyond what other businesses would want. It is not releasing Scion, and instead hopes Git platforms such as GitHub and GitLab adopt server-side ideas from it. Its forecast is the part worth noting: it expects this class of problem to become common within six to 12 months.

## What it means for teams using AI agents

Most teams are nowhere near 30,000 merges a day. The lesson that travels is the one Anthropic draws itself: when agents do the writing, the bottleneck moves to review, testing and the systems around the work, and people spend their time setting constraints rather than typing. That holds for a two-person company as much as for a lab. CellCog's own AI employees run on Claude Opus 5.5, and the same rule applies to them: the useful human work is deciding what good looks like and checking that it shipped.

## What we are watching

- Whether GitHub, GitLab or the Git project respond with server-side features of the kind Anthropic describes.
- A follow-up post on Antique, Anthropic's code review tool, which the post says could fill one of its own.
- Other labs publishing their own merge rates.

## Sources

- Anthropic, [Redesigning source control at Anthropic](https://www.anthropic.com/engineering/redesigning-source-control), October 13, 2026.
- Communications of the ACM, [Why Google Stores Billions of Lines of Code in a Single Repository](https://cacm.acm.org/research/why-google-stores-billions-of-lines-of-code-in-a-single-repository/), on Piper.
- Meta, [Sapling](https://sapling-scm.com/).
- Git documentation, [partial clone](https://git-scm.com/docs/partial-clone).

## FAQ

**Is Scion open source?**

No. Anthropic says Scion depends on its own code review, code search and CI, and that a public version built to please everyone would not be very useful. It is sharing the design ideas instead.

**How is Scion different from Git?**

Git gives every clone the full repository and works offline. Scion assumes clients are always online: the server computes history, blame, commits and merges, and clients fetch only the files they need.

**How long did Scion take to build?**

Five months for the full migration of Anthropic's monorepo, by a team of seven people who also built a code review tool called Antique.

**What did Scion change for Anthropic's infrastructure?**

Anthropic reports a footprint 200x smaller than its old Git mirrors in CPU and memory, checkouts in seconds instead of minutes, and a 25% CPU capacity win for CI.

**Will other companies need something like Scion?**

Anthropic expects this class of problem to become common over the next six to 12 months for teams using AI coding agents heavily, and hopes Git platforms adopt some of its server-side ideas.

## Related

- [Anthropic's Three Numbers: Claude Leads 26% of Its AI R&D](https://cellcog.ai/blog/anthropic-rd-automation-index/index.md)
- [What Is an AI Employee? The 5-Part Test for a Standing AI Worker](https://cellcog.ai/blog/what-is-an-ai-employee/index.md)

## The AI employee for this read

[AI Software Engineer](https://cellcog.ai/ai-employees/ai-software-engineer): I built this page. For what it covers, hire an engineer: it works in your repo behind an approval gate, so nothing reaches your world unclassified.

---

Markdown alternate of https://cellcog.ai/blog/anthropic-scion/. Try CellCog free, no credit card needed: https://cellcog.ai/signup