Anthropic has implemented a bespoke source control system and a code-review tool after a substantial increase in the rate of code changes at the company. According to published figures, in May 2026 Anthropic merged roughly 2,500 changes per day on average; by September 2026 that number exceeded 30,000 merges per day.
Reasons for the change
Anthropic set extreme scalability requirements for their new tooling. The history of their main branch alone is almost 100 GB, and a compressed working copy is over 25 GB. Their continuous integration (CI) infrastructure performs on the order of 25 million code checkouts per day; that count excludes developer environments, agent sandboxes, and research jobs. Naively handled in Git, that traffic would amount to over 600 petabytes per day.
To meet these needs the company developed a new source control system named Scion and a code-review system called Antique. Observers note that Scion’s design resembles a centralized version-control approach in some respects, but it was tailored to Anthropic’s specific operational requirements.
Development and migration
A team of seven people built the new system, and the migration from Git was completed in five months. Anthropic published charts documenting the rise in daily merged changes over the period.
Visualization and model release markers
The article’s author reports having Claude (Anthropic’s model) edit the company’s chart to add markers for public release dates of Anthropic’s 2026 models. Those markers indicate public release dates rather than the dates models became available for internal use. The author speculates that the introduction of Fable/Mythos-class models may be a key factor in the increased coding velocity.
Use of adversarial review
Anthropic also makes extensive use of adversarial review. While the practice is effective in many respects, it can be myopic: the company found that skilled engineers were needed to prevent Claude from driving over-engineering of parts of the system in response to hypothetical concerns. The article’s author notes similar experiences when running adversarial review in their own projects.
Why this matters
The case illustrates how the development and deployment of large generative models can reshape software development workflows and infrastructure needs. Organizations with very high rates of frequent changes may opt to build custom tools and architectures to address scaling and performance constraints.
Summary
Anthropic created Scion and Antique to handle sharply increased code-change throughput. The seven-person development team completed migration from Git in five months as daily merged changes rose from about 2,500 in May 2026 to over 30,000 by September 2026.
