Observed facts: Anthropic reports about 2,500 merged code changes per day in May 2026 and more than 30,000 per day by September. Its main-branch history is nearly 100 GB, its compressed working copy exceeds 25 GB, and CI performs around 25 million checkouts daily. The seven-person team completed the Git migration in five months.
Mechanism: Scion centralizes source control around Anthropic’s unusually heavy usage rather than treating Git compatibility and offline work as universal requirements. The company says naive fetching at its checkout volume would exceed 600 PB of traffic per day. Its account says offline support was not prioritized because Claude, its heaviest Scion user, needs to be online. Why now: The reported increase from roughly 2,500 to more than 30,000 daily merged changes gives Anthropic a direct reason to revisit infrastructure that had previously served a lower-throughput workflow.
First implication: Agent-assisted engineering can shift bottlenecks from producing code toward the systems that store, distribute, review, and validate rapidly growing change volumes. Evidence boundary: These figures and design choices are Anthropic’s account, relayed in Simon Willison’s post; they demonstrate a company-specific response, not that Scion would outperform Git for other organizations.
First watch condition: Watch whether Anthropic reports that Scion’s design measurably improves checkout costs, developer throughput, or review quality beyond its own environment. The account also describes extensive adversarial review and warns that it can lead engineers toward over-engineering in response to hypothetical concerns. That is a reminder that added review capacity needs human steering, not only more automated scrutiny.