TL;DR
Get business pricing on tech for your team
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
Anthropic says it built Scion, an in-house version control system, and migrated its monorepo from Git in five months. The company reports that daily merged changes rose from about 2,500 in May 2026 to more than 30,000 in September, putting pressure on infrastructure designed for smaller, human-led workloads.
Anthropic says it built Scion, an in-house version control system, and moved its monorepo off Git in five months as AI-assisted coding drove a steep rise in internal software changes. The company reported merging more than 30,000 code changes per day in September 2026, compared with an average of about 2,500 per day in May, and said its existing developer infrastructure was becoming a constraint.
In a report published by Anthropic, the company said the shift was part of a broader effort to rebuild infrastructure for a workload increasingly involving AI agents and automated jobs. Alongside Scion, Anthropic says it developed a code-review interface called Antique, code search, an internal source-control permissions system, and tools for reviewing and landing large volumes of changes. The report focuses mainly on Scion.
Anthropic described its repository as unusually large. It said the history of its main branch is almost 100 GB, while a compressed working copy exceeds 25 GB. The company also reported about 25 million code checkouts per day for continuous integration, excluding developer environments, agent sandboxes and research jobs. It estimated that fetching the repository naively at that volume would create more than 600 petabytes of Git traffic per day; this is a projection, not a report of traffic actually incurred.
Before building Scion, Anthropic said it used Git scaling techniques including shallow clones, sparse checkouts, mirrors and repository snapshots stored in blob storage. At its peak, the company operated several hundred internal Git mirrors, according to the report. Anthropic said the growing merge rate made those systems harder to maintain, while snapshots fell further behind the main branch and partial clones complicated operations that rely on history, such as diff and blame.
Why Anthropic Changed Its Source Control
The change reflects a shift in what Anthropic says limits its software development: not the production of code, but the systems needed to build, test and manage it. If AI tools and agents generate changes faster than existing systems can review, merge and check them, engineering throughput can be limited by infrastructure rather than by the act of writing code.
Scion also represents a trade-off from Git’s decentralized model. Git clients generally hold repository data and can work offline; Anthropic designed Scion for clients it says are usually connected, with a server handling more operations and clients keeping a bounded local cache. This can reduce the data each client needs, but relies more heavily on a working central service and network access. Anthropic says its design suits its agent-heavy workload; it does not establish that the same approach would benefit every organization.
The report may be relevant to companies scaling AI-assisted development because it describes infrastructure pressure beyond code generation itself. Anthropic argues that API access to source-control queries can also support automated tools without requiring each job to maintain a full local repository. The scale and cost of its systems, however, are specific to the company’s own environment.
As an affiliate, we earn on qualifying purchases.
Git’s Limits in Anthropic’s Monorepo
Git is a distributed version control system: a clone can contain the repository’s history and support many operations offline. That model works well across a wide range of projects, but large monorepos can make complete copies costly to transfer and maintain. Git includes ways to fetch less data, and organizations often add internal mirrors or prepared snapshots to reduce repeated downloads.
Anthropic says it had adopted those measures but found that its combination of repository size, checkout volume and rising merge rate made them increasingly difficult to operate. The company also points to earlier large-scale source-control systems, including Google’s Piper and Meta’s Sapling, as prior work it considered while designing Scion. Anthropic says it completed the migration in five months; the report does not provide an independent evaluation of the system.
“The bottleneck, now, is all of the developer infrastructure built for human scale: source control, builds, and CI.”
— Anthropic, in its engineering report
large code repository management tools
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
What Anthropic Has Not Detailed
The account is Anthropic’s own report; it does not include an independent audit of Scion’s performance, reliability, security or operating costs. The company gives figures for its code-change rate and repository workload but does not publish comparable before-and-after measurements showing how much Scion reduced checkout times, infrastructure spending or developer delays.
It is also unclear from the report how Scion behaves during outages or network interruptions, what safeguards govern server-side operations, and how the system handles access for people or tools that need offline work. Anthropic’s estimate of more than 600 petabytes of daily Git traffic describes what naive fetching would imply, not a measured volume. The company has not established that its internal results will transfer to organizations with different repository sizes or development practices.
software development infrastructure tools
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Scion’s Results and Wider Use
Anthropic’s report describes the system and its design rationale, but does not announce a public release or give a schedule for one. The next useful evidence will be operational detail: how Scion performs under sustained loads, how it affects merge and review times, and what resources it requires compared with the Git-based setup it replaced.
For now, Anthropic presents Scion as an internal response to its own growth in AI-assisted software work. Whether other organizations adopt similar server-centered source control will depend on their workloads and on evidence about reliability, costs and usability that the report does not yet provide.
As an affiliate, we earn on qualifying purchases.
Key Questions
What is Scion?
Scion is Anthropic’s in-house version control system, built to manage the company’s monorepo. Anthropic says its server handles operations such as answering history and blame queries, creating commits, and performing merges and rebases, while clients keep a bounded local object cache.
Why did Anthropic move away from Git?
Anthropic says its repository and checkout workload had grown beyond what its Git-based setup could comfortably support. It reported more than 30,000 merged changes per day in September 2026 and about 25 million daily CI checkouts, and said operating hundreds of mirrors had become harder as the merge rate rose.
How long did the migration take?
Anthropic says it built Scion and migrated its monorepo from Git in five months. The report does not give a more detailed migration timeline.
Does Anthropic’s report show that Scion is faster or cheaper?
Not with published comparative measurements. Anthropic explains why it built the system and describes the workload it is designed for, but the report does not provide before-and-after figures for speed, cost or reliability.
Will Scion be available to other companies?
The report does not announce a public release or say that Scion will be offered to other organizations. Anthropic describes it as an internal system developed for its own monorepo and agent-heavy workload.
Source: hn
Halloween Picks
halloween
As an affiliate, we earn on qualifying purchases.
