Anthropic Redesigns Source Control for Agent-Scale Coding

Summary

Anthropic says the rapid growth of Claude-assisted coding made human-scale developer infrastructure, especially Git-based source control, its new bottleneck. Daily merged changes rose from about 2,500 in May 2026 to more than 30,000 by September, while the monorepo’s main-branch history approached 100 GB, its compressed working copy exceeded 25 GB, and CI performed roughly 25 million checkouts per day. The company built Scion, an in-house source control system, and migrated its monorepo from Git in five months alongside a code-review tool called Antique, code search, and access-control systems. Scion assumes clients such as agents and jobs are always online, so the server performs history queries, commits, merges, and rebases while clients keep only bounded caches and sparse or virtualized working copies. Changesets store only modified files and content-addressed blob pointers; derived structures provide blame, path history, commit-graph, search, and copy-detection functionality without requiring a fully materialized repository. Its tiered server, cache, and node-daemon architecture uses immutable objects and horizontal scaling, with cache hit rates above 99%. Anthropic reports that Scion’s infrastructure footprint is 200 times smaller than its former Git-mirroring setup, checkouts are two orders of magnitude faster, and CI gained 25% CPU capacity. A seven-person team built Scion and Antique and completed the migration in five months, using Claude for implementation and optimization while engineers retained responsibility for system design, constraints, and review intent. Anthropic says Scion is not a general replacement for Git because it requires major ecosystem migration and bespoke tooling, but argues that its server-side APIs, sparse access, and agent-oriented architecture could inform future Git platforms as agentic development grows.