Anthropic Replaced Git with Scion System for AI Agent Code

Anthropic published a research paper on its proprietary version control system, Scion, which replaced Git for Claude agent code in 5 months by 7 engineers as daily merges grew from 2,500 to 30,000.

Compiled by Sergey KostenchukPublished 2026-10-20Updated 2026-10-20

2026-10-20 Coding Anthropic

💻 Anthropic Replaced Git with Scion System for Claude Agent Code

The migration from Git was completed by 7 engineers in 5 months, while also building the review tool Antique, code search, and IAM. The reason was the growth in the flow of changes from Claude agents: about 2,500 merges per day in May 2026, and more than 30,000 by September, with 25 million CI checkouts per day.

🌍 For the first time, it is shown how a VCS for AI agents, not humans, is structured: server-side commits, API instead of local blame and diff, and distributed caching. Scion requires 200 times less CPU and memory than Git mirroring, checkout speed increased by two orders of magnitude, and CI saved 25% of compute resources.

👤 The article includes an analysis of real-world solutions: changeset as a WAL record, sparse/virtualized checkout, and comparison with Git packfiles. Useful for engineers whose code is increasingly written by agents.

Source 1: https://www.anthropic.com/research/redesigning-source-control Source 2: https://news.ycombinator.com/item?id=50171167

Sources