Anthropic’s Scion migration is a concrete, unusually large-scale example of that shift.
Observed facts: Anthropic reports that its daily merged changes rose from about 2,500 in May 2026 to more than 30,000 in September. It says it migrated its monorepository from Git to Scion in five months. The company also describes roughly 25 million code checkouts per day for CI and a repository history approaching 100 GB.
Mechanism: Scion assumes clients are online and makes the server responsible for history queries, commits, merges, and rebases, while clients keep bounded caches and request data as needed. Its APIs allow jobs to query source-control information without maintaining full local repositories; sparse and virtualized checkouts avoid fetching files that a job does not need. The mechanism is to move repository work behind scalable server APIs and make clients fetch only the data their task needs.
Why now: The reported jump in merge volume makes infrastructure built around human-scale repository workflows a visible constraint. Anthropic frames the underlying driver as Claude’s ability to write and iterate on software at much higher volume. The first implication is that teams adopting coding agents may need to measure and improve merge, checkout, build, and review capacity—not only coding speed.
Evidence boundary: This is Anthropic’s account of its own workload and system; its scale is explicitly unusual. The post is a case study, not comparative evidence that Scion or a centralized design is preferable for ordinary repositories. The evidence does not show that most teams need to replace Git now.
What to watch: Watch whether other high-throughput teams report repository, CI, or review infrastructure becoming a measurable bottleneck as agent-generated changes increase. Anthropic’s account is useful as an early operational signal, but broader adoption and independent performance comparisons remain unestablished.