The thesis is: *For production agents, the harness is becoming a more durable product boundary than the identity of the model running inside it.* Anthropic reports about 2,500 merged changes per day in May 2026 and more than 30,000 in September. It also reports around 25 million CI checkouts each day, excluding other environments.
The mechanism is architectural, not simply a faster Git client. *Scion assumes clients are online and moves operations such as history queries, commits, merges, and rebases to a server.* The client maintains a bounded cache and materializes files as needed; APIs let tools ask source-control questions without holding a complete repository. Sparse and virtualized working copies reduce unnecessary checkout work. The implication is: *When agent-driven code production scales, teams may need to treat source control, CI, and review as a coordinated execution harness rather than separate developer utilities.* This is a company-authored account of Anthropic’s own unusually large workload, not evidence that typical teams need a bespoke version-control system.
Why this matters now: *Anthropic’s reported increase in merge volume makes infrastructure limits visible as a near-term consequence of coding-agent adoption.* The evidence boundary is: *The post provides Anthropic’s reported workload and design rationale, but no independent benchmark or evidence that Scion’s approach generalizes to ordinary repositories.* What to watch: *Whether other organizations report similar pressure on checkout, CI, and source-control services as agent-generated changes increase.* The practical question is not whether every team should replace Git, but whether its existing tooling can absorb changes at the rate its agents create them.