Anthropic reinvents source control as Claude floods the world with code
Anthropic has built Scion, a source control system designed to handle the volume of code produced with help from its Claude models. The shift matters because the company says managing, reviewing and building code has become a bigger constraint than writing it, as automated coding drives a rapid rise in changes.
Anthropic said it merged about 2,500 code changes a day in May 2026 and more than 30,000 a day by September. Its main branch is around 100 GB, and it runs about 25 million code checkouts daily for continuous integration; fetching those naively would use more than 600 petabytes of Git traffic per day. Scion uses a central server and records changes as changesets, while Claude also helped develop related review and search tools; Anthropic says an internally tested rewrite of an ingestion pipeline used 20 times less memory and ran four times faster.
- Anthropic says daily code merges rose from 2,500 to over 30,000.
- Scion is its custom source control system for a large monorepo.
- Claude helped rewrite a pipeline to use less memory and run faster.
New here? Start with this
Source control systems keep a record of changes to software, so teams can work on the same project, review updates and return to earlier versions. Git is a widely used system of this kind, and many software teams rely on it to manage their code.
Anthropic makes the Claude family of AI models, which can help write software. As AI tools contribute to more code changes, keeping those changes organised, checking them and running tests can take more effort. Anthropic says this has led it to develop Scion, a system that stores and manages code changes in a different way.
The issue matters to software teams because their tools need to keep pace with the amount of code being produced. Changes to how code is stored and reviewed could affect the speed and resource demands of building software, as well as how teams check that updates work correctly.
Both sides, in good faith
The strongest fair case each way — we don't pick a winner.
The case for
Scion may be a sensible response to a real change in software development: when AI tools sharply increase the volume of code, systems built around older workloads can make review, integration and testing the bottleneck. A central service designed for frequent changes and large repositories could reduce wasted computing and help teams deliver improvements faster, while Anthropic’s reported pipeline results suggest the approach merits serious testing.
The case against
A company’s internal results do not yet show that a new, centrally managed system will work as well across different teams, repositories and workflows, or that the claimed gains will outweigh migration and maintenance costs. Source control is foundational infrastructure, so replacing familiar, widely used tools could introduce dependence on one provider and create risks around access, resilience and control over code; teams may reasonably prefer to improve existing tools until Scion’s benefits are independently demonstrated.