Anthropic details Scion, a git replacement for its monorepo, built by seven engineers with Claude in five months, as daily merges rose from 2.5K in May to 30K+

Your code merges become materially faster, particularly if you use AI assistants to generate code.

Anthropic has developed "Scion," a custom version control system designed to replace Git for its massive monorepo. This initiative was driven by a tenfold increase in daily code merges, surging from 2,500 in May to over 30,000 by September 2026, which had become a significant bottleneck. The system was built by a team of seven engineers, leveraging Anthropic's own AI, Claude, over a five-month period. Scion is built on principles adapted for agent-scale development, assuming clients are always online and implementing a client-server asymmetry where the server handles complex operations. Unlike Git, which assumes offline, distributed clients, Scion's server-centric architecture enables horizontally scalable components and avoids the need for full repository materialization. Additional tools like "Antique" for code review and enhanced code search were also developed concurrently.

Anthropic's development of Scion signals a critical inflection point for software development tools as AI agents dramatically increase code output. The shift from Git, designed for human-scale, distributed workflows, to a server-centric, always-online model is a necessary adaptation for managing the unprecedented scale of AI-driven development. This move suggests that the traditional assumptions underpinning widely adopted developer tools may soon become obsolete for leading AI-native organizations, prompting a potential wave of specialized infrastructure development.

Companies heavily investing in AI for code generation and those operating at a similar scale to Anthropic may soon face similar source control bottlenecks. The success of Scion could encourage the development of other specialized, AI-aware version control systems or lead existing players to fundamentally re-architect their offerings. Organizations should monitor how these new systems handle scale, agent interaction, and the integration of AI within the development lifecycle, as these advancements directly impact developer productivity and deployment speed.