AIThis post was created with the assistance of artificial intelligence (AI).

TL;DR

Buying for a business?Offer from Amazon

Get business pricing on tech for your team

  • Business-only prices and quantity discounts
  • Tax-exempt purchasing
  • Multiple users, one account, clear invoices
As an affiliate, we earn on qualifying purchases.

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 code changes rose from about 2,500 in May 2026 to more than 30,000 in September, putting pressure on infrastructure designed around smaller, human-scale workloads.

Anthropic says it has migrated its monorepo from Git to Scion, an in-house version control system built over five months, after its daily merged code changes climbed from about 2,500 in May 2026 to more than 30,000 in September. The company says the shift is intended to handle a workload increasingly driven by Claude-assisted coding and to reduce pressure on infrastructure designed for smaller repositories and human-scale development.

In a report published by Anthropic, the company describes Scion as a system designed for clients that are online by default, including AI agents, continuous integration jobs and research workloads. Unlike Git, which gives each clone a full repository and supports offline work, Scion relies on a server for operations such as querying history, creating commits, merging and rebasing. Its client keeps a bounded local object cache and manages working files.

Anthropic says the scale of its repository had made its earlier Git-based setup increasingly difficult to operate. It reports that the main branch’s history is nearly 100 GB, while a compressed working copy exceeds 25 GB. The company also says its CI system performs about 25 million code checkouts per day, before counting developer environments, agent sandboxes and research jobs.

Before building Scion, Anthropic used Git tools such as shallow clones and sparse checkouts, as well as internal mirrors and repository snapshots. The company says it operated several hundred Git mirrors at peak, but that increasing merge volume made them harder to maintain. Alongside Scion, Anthropic says it developed a code-review interface called Antique, code search and an internal source-control access system; the report focuses mainly on Scion.

At a glance
reportWhen: Reported in September 2026; Anthropic s…
The developmentAnthropic says it created Scion and moved its monorepo from Git to the new system in five months to support a sharp rise in code changes and repository activity.

Why Anthropic Reworked Source Control

The report offers a view of how AI-assisted software production can shift pressure from writing code to the systems that store, test and review it. Anthropic says removing one infrastructure bottleneck can lead to more code being produced, which in turn places more demand on the next parts of the development pipeline. That makes source control a question of capacity and workflow design, not just developer convenience.

Scion’s server-centered approach reflects a specific trade-off: agents and automated jobs can request repository data as needed instead of each keeping a full copy. Anthropic says exposing source-control queries through an API also lets internal tools access history and other information without maintaining a local repository. Whether that design suits other organizations will depend on their own repository size, network reliability and need for offline work.

Amazon

version control system for large codebases

As an affiliate, we earn on qualifying purchases.

As an affiliate, we earn on qualifying purchases.

From Git Mirrors to Scion

Git’s distributed model is useful because every clone can hold the repository’s history and perform many operations without a server connection. Anthropic says that design was a poor fit for its particular workload as its monorepo and checkout volume grew. The company estimates that naively fetching a full copy for its roughly 25 million daily CI checkouts would amount to more than 600 petabytes of Git traffic per day; it presents this as a scale illustration, not as measured traffic.

Anthropic says it first relied on established Git scaling techniques, including mirrors and stored repository seeds. As its merge rate rose, seeds fell further behind the main branch between refreshes, while partial clones required extra care for history-dependent tasks such as diff and blame. The company says its redesign drew on prior work on large-scale source control, including Google’s Piper and Meta’s Sapling. The report does not present Scion as a general replacement for Git, but as a system shaped around Anthropic’s online, automated workload.

“The bottleneck, now, is all of the developer infrastructure built for human scale: source control, builds, and CI.”

— Anthropic, in its engineering report

Amazon

enterprise code review tools

As an affiliate, we earn on qualifying purchases.

As an affiliate, we earn on qualifying purchases.

What the Report Leaves Open

The report describes Anthropic’s architecture and workload, but does not provide independent performance results or a detailed comparison of Scion’s costs, reliability and operational complexity against the former Git setup. It is also unclear how the system behaves during service interruptions, how much repository data clients typically cache, or what safeguards and recovery processes support the centralized services.

Anthropic says its workload is unusual and does not claim that most organizations need to move away from Git now. The company expects some organizations using AI coding tools could encounter similar pressures, but the timing and scale of that shift remain uncertain. The reported merge figures and system metrics are company-reported; the source material does not include an outside audit.

Amazon

software development repository management

As an affiliate, we earn on qualifying purchases.

As an affiliate, we earn on qualifying purchases.

How Scion’s Use May Develop

Anthropic says it does not expect its rate of code production to slow soon, and its report presents Scion as part of a wider effort to scale source control, builds and review. The next practical test will be whether the system continues to support increasing repository activity while keeping automated and human workflows dependable.

The report does not announce a public release of Scion or give a timetable for sharing it beyond Anthropic. Further details on operating results, reliability and the performance of its associated tools would help readers judge how broadly the approach may apply. For now, the confirmed development is an internal migration, not a change to Git’s status as a general-purpose version control system.

Amazon

AI-assisted coding tools

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. The company says its server performs repository operations while clients keep a bounded local cache and manage working files.

Why did Anthropic move its monorepo away from Git?

Anthropic says its growing repository and automated workload made Git mirrors and other scaling measures harder to operate. It designed Scion for clients such as agents and CI jobs that are online and can request data from a server.

How much did Anthropic’s merge rate increase?

Anthropic reports about 2,500 merged code changes per day in May 2026 and more than 30,000 per day in September 2026. These are company-reported figures for those periods.

Does Anthropic say other companies should replace Git?

No. The report describes Scion as a response to Anthropic’s own unusually large, automated workload. The company suggests similar pressures may reach other AI-heavy development environments, but does not recommend a general move away from Git.

Is Scion available outside Anthropic?

The report does not announce a public release or provide a timetable for making Scion available externally. It describes the system as an internal tool used for Anthropic’s monorepo.

Source: hn

HALLOWEEN

Halloween Picks

As an affiliate, we earn on qualifying purchases.

You May Also Like

Test Data Management Becomes a Problem Earlier Than You Think

Test data management challenges arise sooner than expected due to increasing complexity and manual processes, but automation offers effective solutions to stay ahead.

Future of Open-Source Development

Navigating the future of open-source development reveals transformative licensing and governance trends that will shape how you contribute and innovate moving forward.

Infrastructure Drift Is a Bigger Risk Than Most Teams Notice

AIThis post was created with the assistance of artificial intelligence (AI).Infrastructure drift…

Disk Is the Contract: Inside Threlmark’s Local-First Architecture

Threlmark treats disk storage as the definitive source of truth, simplifying sync, enhancing offline use, and enabling data portability without traditional databases.