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 source control system, and migrated its monorepo from Git in five months. The company reported more than 30,000 code changes merged per day in September 2026, up from an average of about 2,500 per day in May, and said its online agent-heavy workload strained its previous Git infrastructure.
Anthropic says it migrated its monorepo from Git to a homegrown version control system called Scion in five months, responding to a sharp rise in code changes produced and merged with help from AI coding tools. The company reported an average of about 2,500 merged changes per day in May 2026 and more than 30,000 per day by September, putting pressure on infrastructure built around conventional developer workflows.
In a report published by Anthropic, the company said its main branch history had grown to nearly 100 gigabytes, while a compressed working copy exceeded 25 GB. It also reported about 25 million code checkouts per day for continuous integration alone, not counting developer environments, agent sandboxes, or research jobs. Anthropic estimated that naively fetching repository data for that workload would amount to more than 600 petabytes of Git traffic per day.
Anthropic said it had tried familiar ways to manage large Git repositories, including shallow clones, sparse checkouts, internal mirrors, and repository snapshots stored in blob storage. At its peak, the company operated several hundred Git mirrors, but said increasing merge volume made them harder to maintain and left snapshots further behind the main branch between refreshes. Partial clones also complicated history-dependent operations such as diff and blame, according to the company.
Scion changes the division of work between client and server. Rather than keeping a full repository locally, the client holds a bounded object cache and requests information from an always-available server. Anthropic says the server handles operations including history and blame queries, commit creation, merges, and rebases. Its source control queries are also exposed through an API, which the company says can let CI jobs and internal tools work without maintaining a local repository.
Why Anthropic Shifted Work to Servers
The change reflects a pressure point created when software development tools are used by large numbers of automated agents: the volume of code activity can outgrow infrastructure designed for people. Git’s distributed model lets each clone operate independently and offline, a strength for many teams. Anthropic says that model was increasingly costly for its own monorepo because many of its clients are online agents and jobs that repeatedly need repository data.
Scion’s server-centered design could reduce the amount of repository data that each automated job must fetch and keep locally. Anthropic also says making source-control operations available through an API supports internal services such as automated bisect and revert pipelines. Those are the company’s reported goals and experiences; the post does not provide independent measurements of Scion’s performance, operating costs, or reliability compared with its previous setup.
The report may be relevant to other organizations increasing their use of AI coding tools, but Anthropic cautions that its workload is unusual and does not say that most companies currently need to replace Git. Its account instead offers an example of how infrastructure demands may change as coding agents generate and process more changes.
enterprise source control management tools
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
From Git Mirrors to Scion
Git is a decentralized version control system: a clone can contain repository history and support operations such as viewing diffs or checking out revisions without a server connection. Anthropic describes that design as appropriate for most repositories, but says it became a poor fit for a monorepo with a very large history and an unusually high rate of checkouts.
The company said it had used established Git scaling techniques before building Scion. It also pointed to prior systems for large monorepos, including Google’s Piper and Meta’s Sapling, as examples whose experience informed its approach. Scion was part of a wider internal tooling effort: Anthropic says that during the same five-month period it also built a code review interface called Antique, code search, and an access-control system for source control.
“The bottleneck, now, is all of the developer infrastructure built for human scale: source control, builds, and CI.”
— Anthropic, in its engineering report
large repository management software
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
What Anthropic Has Not Disclosed
The report does not give independent verification of the daily merge figures or detailed benchmarks comparing Scion with Git. It also does not specify Scion’s full deployment architecture, its infrastructure costs, how it handles outages or offline work, or whether the system will be released outside Anthropic. The company describes its design and operational experience, but does not publish enough data to assess its performance across other kinds of repositories.
Anthropic’s projection that code production will continue growing is the company’s expectation, not a confirmed forecast. It is also unclear whether other organizations using coding agents will encounter the same scale of repository traffic or find a server-dependent design suitable for their security, availability, and workflow needs.
AI developer tools for code collaboration
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
How Scion’s Use May Evolve
Anthropic’s report describes Scion as an operating system for its monorepo, but does not announce a public release or a timetable for sharing the software. The next developments to watch are whether the company publishes further technical details or performance data, and whether it expands on how Scion supports code review, CI, and automated repository operations.
For other engineering teams, the practical next step is not necessarily to replace Git. Anthropic’s account points to a narrower question: whether rising agent-driven checkouts and merges are straining existing systems, and whether incremental measures such as repository mirrors remain manageable at that scale. The report does not set a threshold for when a different architecture would be warranted.
server-based version control systems
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Key Questions
What is Scion?
Scion is Anthropic’s in-house version control system, built for the company’s monorepo. Its clients request repository data from a server rather than keeping a complete local copy.
Why did Anthropic move away from Git for its monorepo?
Anthropic says its large repository and high volume of automated checkouts made Git mirrors and partial-clone workflows harder to operate. The company designed Scion around the fact that most of its coding agents and jobs are online.
How many code changes did Anthropic say it merged?
Anthropic reported an average of about 2,500 merged changes per day in May 2026 and more than 30,000 per day by September 2026. These are figures reported by the company.
Is Scion available to other companies?
The report does not announce a public release or say that other companies can use Scion. Anthropic presents it as an internal system and describes its experience as unusually demanding.
Does Anthropic say other teams should stop using Git?
No. Anthropic says Git remains the right design for most repositories and cautions that many organizations are unlikely to face its current workload in the near term.
Source: hn
Halloween Picks
halloween
As an affiliate, we earn on qualifying purchases.
