Anthropic redesigns source control with Scion for agent-scale monorepos
Anthropic 的这篇 《Redesigning source control at Anthropic》 很值得一看： https://t.co/Ah6MIcHT74 它们遇到的问题是： 5 ...
The deeper signal is that AI agents writing code at scale break assumptions Git was built on, and Anthropic had to redesign infrastructure around that reality. Engineering leaders planning monorepo or agent-driven CI workflows can study the server-heavy, API-first architecture as a blueprint. It also hints that similar scaling pressure will hit any team adopting large-scale coding agents.
Anthropic published an engineering post titled 'Redesigning source control at Anthropic' describing Scion, a new source control system. Merges grew from about 2,500 per day in May to over 30,000 per day by September, and the monorepo reached 100 GB with roughly 25 million CI checkouts per day. Scion is not a Git replacement; it uses a client-server asymmetric design where the server handles heavy work while clients keep a limited object cache and working copy. Data is sharded across multiple machines so no single machine stores the full repository, and all queries go through an API, letting many CI jobs and tools run without a local checkout.
Anthropic 的这篇 《Redesigning source control at Anthropic》 很值得一看： https://t.co/Ah6MIcHT74 它们遇到的问题是： 5 ...
Anthropic 的这篇 《Redesigning source control at Anthropic》 很值得一看： anthropic.com/engineering/re… 它们遇到的问题是： 5 月平均每天合并约 2 500 次，到 9 月已超过每天 30000 次 单仓库 monorepo 规模太大，有 100 G ，每天仅 CI 就要做约 2500 万次 checkout，流量非常大 所以重新设计了 Scion，不是 Git 的替代，适合超大 monorepo + 大量 Agent 写代码的场景 客户端不用自己存完整仓库 客户端-服务器不对称设计，服务器干重活，客户端维护有限对象缓存，工作副本等等 支持水平扩展，数据分散存在多台机器上，没有任何一台机器需要存下完整仓库。封装了完整的一个源码控制服务，所有查询都通过 API 暴露，很多 CI/工具甚至不需要本地仓库 可以收藏好好看看 💬 2 🔄 0 ❤️ 3 👀 395 📊 3 ⚡
