Anthropic Builds Its Own Version-Control System as AI Pushes Daily Code Merges Past 30,000

When most people think about what slows down a software company, they picture things like hiring freezes, budget cuts, or slow internet connections. They probably don’t picture the humble version-control system — the software that tracks every change a developer makes to a codebase and stitches it all together. But that’s exactly what became a bottleneck at Anthropic, the AI safety company behind the Claude family of models. And the solution the company came up with is worth understanding, because it tells you something real about where AI-assisted software development is heading.

Anthropic’s engineering team has published a detailed account of how its internal code infrastructure buckled under the weight of AI-agent activity, and what it did about it.

From 2,500 to 30,000 Merges a Day

The numbers are striking. According to the Anthropic Engineering blog, the company was averaging around 2,500 merged code changes per day in May 2026. By September 2026, that figure had climbed to more than 30,000 per day — a more than twelvefold increase in roughly four months.

That kind of growth isn’t the result of a hiring spree. It reflects what happens when AI agents start doing a large share of the actual coding work. Where a human developer might produce a handful of changes in a working day, an AI agent can generate many more in the same time. Multiply that across an organisation the size of Anthropic and the infrastructure strain becomes obvious.

The problem, as Anthropic described it, was that its existing developer tools had been designed for human workloads. Git — the open-source version-control system used by the bulk of software teams around the world, from bedroom hobbyists to large enterprises — works well when the people committing changes are, well, people. But it wasn’t built to handle the sheer volume and speed of AI-generated code. Anthropic said its infrastructure had become a genuine bottleneck, constraining how much code could be reviewed, merged, built, and tested.

What Is Git, and Why Does It Matter?

A quick word on Git for anyone who hasn’t encountered it. When software developers write code, they need a way to track what changed, who changed it, and when. Git does that job. It also lets multiple people work on the same project simultaneously without constantly overwriting each other’s work. It’s the backbone of modern software development and is used by millions of teams worldwide.

For most organisations, Git is more than adequate. But Anthropic’s situation had become unusual. The company wasn’t dealing with a few hundred developers making careful, considered changes. It was dealing with AI agents producing code at a pace that Git’s architecture simply wasn’t designed to accommodate.

Scion: Built in Five Months

So Anthropic built its own replacement. The system is called Scion, and it’s described as a change-list-based source-control system built specifically for Anthropic’s internal monorepo — a single large repository that holds the company’s entire codebase. According to the Anthropic Engineering blog, the migration from Git to Scion was completed in five months.

Secondary sources have reported that a small team of around seven engineers built Scion and the associated code-review systems, and that the system moves more processing work to servers rather than local machines, exposing source-control functions through programming interfaces. Those details haven’t been independently verified from the primary blog post, so treat them as reported rather than confirmed.

What Anthropic has confirmed is that removing this bottleneck isn’t expected to reduce demand. If anything, the company says, solving one infrastructure constraint tends to free up capacity that enables even more code production. In other words, 30,000 merges a day may not be the ceiling.

What the Industry Is Watching

This story matters beyond Anthropic’s own walls. It’s a concrete example of something the software industry has been talking about in fairly abstract terms: AI-assisted programming doesn’t just change how code is written, it changes the entire pipeline around it. Review, testing, deployment, and version control all come under new pressure when the volume of code increases steeply.

Git isn’t going anywhere for most teams. Anthropic’s situation is unusual by any measure, and Scion is a proprietary internal system — there’s no indication the company plans to release it publicly or that it’s intended as a replacement for Git in general use. Developers at ordinary companies, from small startups to large enterprises, are unlikely to face these constraints any time soon.

But the broader point stands. As AI agents become more capable and more widely used in software development, the supporting infrastructure around them will need to keep pace. Anthropic’s experience is an early, unusually transparent look at what that challenge actually involves in practice.

What This Means for Kent Residents

There’s no direct impact on Kent residents, local councils, or NHS services from this particular development. For anyone working in software development or technology roles across Kent — whether at a business in Maidstone, a university in Canterbury, or a tech employer along the Thames Gateway — it’s a useful reminder that AI-assisted coding is already reshaping how development teams work at scale, and that the tools and skills surrounding that shift are evolving quickly. For most developers, Git remains perfectly fit for purpose; this is a story about what happens at the very sharp end of AI development, not a signal that everyday tools need replacing.