Anthropic swaps Git for in-house system as AI code surges

Artificial intelligence is changing how software gets built, but producing code faster creates another challenge: managing it efficiently. Anthropic has addressed this problem by replacing Git across its main codebase with Scion, an in-house version control system. The move highlights how AI-generated code is pushing traditional development infrastructure beyond its original limits.

Why Anthropic Replaced Git

Anthropic reported a significant increase in daily code changes. In May 2026, the company merged approximately 2,500 changes per day. By September, that figure had exceeded 30,000, representing more than tenfold growth in just four months.

According to Anthropic, writing code is no longer its biggest constraint. Instead, source control, software builds, and continuous integration (CI) have become bottlenecks. These systems must process, test, and organise the growing volume of changes produced by AI tools.

The company expects demand to continue increasing as AI agents take on more development tasks. Its experience demonstrates why software teams may need to rethink infrastructure designed primarily for human developers.

What Is Anthropic Scion?

Anthropic Scion is an in-house version control system designed to manage an exceptionally large codebase and support high volumes of automated development activity.

Git uses a decentralised approach that allows developers to maintain complete repository copies, track changes, and work offline. This model works well for most projects, but Anthropic’s workload presents different challenges.

The company’s main repository has a history approaching 100 GB, while a compressed working copy exceeds 25 GB. Anthropic also performs approximately 25 million code checkouts daily for CI.

Downloading repository data inefficiently at that scale could generate more than 600 petabytes of Git traffic per day. Before switching, Anthropic operated several hundred internal Git mirrors to handle the demand.

Scion takes another approach by assuming that most clients, including AI agents and automated jobs, remain online. Its server handles operations such as commits, merges, rebases, and file-history queries, reducing the work required on individual machines.

How Scion Improves Development Performance

One of Scion’s main advantages is its ability to avoid unnecessary data transfers. Instead of downloading an entire repository, automated jobs can use sparse or virtualised checkouts to access only the files they need.

This approach helps Anthropic reduce computing demands while keeping development workflows moving.

The company reports that Scion uses 200 times less CPU and memory than its previous Git mirror infrastructure. Checkouts that once took minutes now finish in seconds, and CI has recovered 25% of its CPU capacity.

These improvements matter because testing and reviewing code can become expensive when thousands of changes arrive daily. Faster checkouts allow automated jobs to access necessary files more quickly, while reduced infrastructure overhead leaves additional capacity for other tasks.

However, these results reflect Anthropic’s unusually large workload. Smaller development teams may not need a specialised version control system to achieve efficient results.

Claude Helped Build the New System

Anthropic developed Scion with a team of seven engineers, alongside a code review tool called Antique and several related systems. The team completed the migration in five months while daily merge volumes increased dramatically.

Claude also contributed to building the infrastructure. Anthropic described an example in which an early internal version of Claude Mythos Preview rewrote a pipeline that transferred data from Git into Scion.

The rewritten implementation used 20 times less memory and ran four times faster than the earlier version, according to the company.

Despite these gains, human engineers remained responsible for architectural decisions and reviewing proposed changes. Anthropic explained that their work increasingly involved directing Claude, evaluating its solutions, and rejecting unnecessary complexity.

This demonstrates how AI can help develop the infrastructure needed to support further AI adoption, while human judgment remains important for maintaining quality and reliability.

Will Anthropic Scion Replace Git?

Anthropic does not intend to release Scion publicly or position it as a general replacement for Git. The system is designed for enormous monorepos and highly automated workloads rather than everyday software projects.

Instead, the company plans to share its engineering principles and work with GitHub, GitLab, and the open-source community to explore improvements to existing tools.

Anthropic expects similar scaling challenges to emerge at other organisations within six to twelve months as AI-assisted development expands.

For businesses, the key lesson is to evaluate the entire software delivery process. Monitoring checkout times, build performance, testing queues, and merge delays can help teams identify infrastructure problems before they affect productivity.

Call to action: If your organisation uses AI coding assistants, review your development pipeline and identify bottlenecks before rising code volumes slow your team down.

FAQs

1. What is Anthropic Scion?

Anthropic Scion is an in-house version control system developed to manage Anthropic’s large repository and handle the increasing volume of AI-generated code changes.

2. Why did Anthropic replace Git with Scion?

Anthropic experienced a surge in daily code merges and approximately 25 million CI checkouts per day. Scion reduces computing overhead and makes repository checkouts significantly faster.

3. Is Anthropic Scion available to the public?

No. Anthropic does not plan to release Scion publicly because it is designed for exceptionally large repositories and automated workloads. The company intends to share its design ideas with the developer community.

Conclusion

Anthropic’s switch to Scion shows how AI-generated code is reshaping software development infrastructure. With daily merges increasing from around 2,500 to more than 30,000, Anthropic needed a system capable of handling its growing workload. Although Scion is not intended to replace Git for everyone, its performance improvements highlight the importance of scalable development tools as AI adoption accelerates.