{"slug": "scion-how-anthropic-replaced-git-with-centralized-source-control-for-30000-agent", "title": "Scion: How Anthropic Replaced Git with Centralized Source Control for 30,000 Agent-Generated Commits Per Day", "summary": "Anthropic replaced Git with a custom centralized version control system called Scion after its commit rate jumped from 2,500 to 30,000 per day following internal deployment of Fable and Mythos-class models. Built by a seven-person team and migrated in five months, Scion uses server-side derived data and partial checkouts to cut checkout cost from O(history size) to O(changed files), dropping offline support that agents never needed. The company also built a custom code review system, Antique, which uses adversarial review where Claude critiques its own changes, though engineers noted the agent tends to over-engineer in response to hypothetical concerns.", "body_md": "Anthropic replaced Git with a custom centralized version control system called Scion after their commit rate jumped from 2,500 to 30,000 per day. The spike correlates directly with internal deployment of Fable and Mythos-class models. Claude is now the heaviest user of their source control infrastructure, and the system was explicitly designed around the assumption that agents are always online.\n\nThis is not a story about better tooling. It is a story about how agent-generated code velocity breaks the assumptions baked into distributed version control.\n\nGit's distributed model assumes developers work offline, clone full repositories, and merge infrequently. Anthropic's workload inverted every assumption:\n\nThe bottleneck was not storage. It was the cost of materializing full working copies for ephemeral compute. Every CI job, every agent sandbox, every research experiment needed a clean checkout. Git's object model forces you to fetch and decompress the entire history graph to get a working tree.\n\nAgents do not care about offline workflows. Claude runs in a data center. It does not need to work on a plane. Anthropic's engineers realized they were paying the complexity cost of distributed version control for a feature their primary user would never need.\n\nScion is a centralized system inspired by Google Piper and Meta's Sapling/Mononoke. The core idea is **derived data**: the server precomputes and caches expensive operations so clients fetch only what they need.\n\n| Component | Git Model | Scion Model | \n|---|---|---|\n| Repository structure | Distributed, full clones | Centralized, partial checkouts | \n| Working copy materialization | Client-side decompression | Server-side derived data | \n| Offline support | First-class | Not supported | \n| Checkout cost | O(history size) | O(changed files) | \n| Primary user | Human developers | AI agents + humans | \n\nThe server maintains precomputed snapshots of every commit. When a client requests a checkout, it fetches only the files it needs for that specific revision. The server handles decompression, delta application, and tree construction.\n\nThis trades server compute for client bandwidth and storage. For Anthropic's workload, that trade is asymmetric: they have centralized infrastructure and need to minimize per-checkout cost.\n\n``` python\n# Conceptual Scion checkout (not actual implementation)\ndef checkout(commit_sha, sparse_paths=None):\n    # Server-side: lookup precomputed tree snapshot\n    snapshot = derived_data_cache.get(commit_sha)\n\n    if sparse_paths:\n        # Only fetch requested subtrees\n        files = snapshot.filter(sparse_paths)\n    else:\n        files = snapshot.all_files()\n\n    # Stream file contents to client\n    for path, blob_id in files:\n        yield (path, blob_store.fetch(blob_id))\n```\n\nClients do not decompress objects. They do not walk the commit graph. They request a snapshot and stream file contents. The server absorbs the complexity.\n\nAnthropic's commit velocity chart shows clear spikes aligned with model releases. The jump from 2,500 to 30,000 commits per day happened after Fable and Mythos became available internally.\n\nAgent-generated commits have different characteristics than human commits:\n\nAnthropic built a custom code review system called Antique to handle this volume. The system uses adversarial review, where Claude critiques its own changes and other models flag potential issues.\n\nThe engineering team noted a specific failure mode: adversarial review can be myopic. Claude over-engineers in response to hypothetical concerns. Human engineers have to steer the agent away from adding complexity for edge cases that will never occur.\n\nThis is a calibration problem. The agent does not know which concerns are real and which are paranoia. It needs human feedback to learn the difference.\n\nA seven-person team built Scion and migrated Anthropic's entire codebase from Git in five months. The migration strategy was:\n\nThe team did not attempt a big-bang migration. They ran both systems in parallel until Scion proved stable under production load.\n\nThe lack of offline support was a conscious trade-off. When asked on Hacker News, a Scion engineer said: \"Since Claude is by far the heaviest user of Scion and is necessarily online, this just wasn't important to do.\"\n\nThis is the clearest signal that Scion is designed for agents first, humans second.\n\nCentralized systems have different failure modes than distributed systems:\n\nAnthropic mitigates these risks with:\n\nThe system is operationally simpler than Git for their workload, but it requires more infrastructure investment. You cannot run Scion on a laptop.\n\n**Use Scion's architecture when:**\n\n**Avoid this approach when:**\n\nAnthropic's migration is a preview of what happens when agents become the primary users of development infrastructure. The tools we built for humans do not scale for agents. Centralized systems with server-side compute are a better fit for workloads where the client is always online and needs fast, cheap checkouts.\n\nThe interesting question is not whether Scion is better than Git. The interesting question is: what other developer tools will need to be redesigned when agents generate more code than humans?", "url": "https://wpnews.pro/news/scion-how-anthropic-replaced-git-with-centralized-source-control-for-30000-agent", "canonical_source": "https://dev.to/mech_app_ai/scion-how-anthropic-replaced-git-with-centralized-source-control-for-30000-agent-generated-5d3e", "published_at": "2026-10-15 10:06:07+00:00", "updated_at": "2026-10-15 10:19:36.933164+00:00", "lang": "en", "topics": ["ai-agents", "developer-tools", "mlops", "ai-infrastructure", "large-language-models"], "entities": ["Anthropic", "Scion", "Git", "Claude", "Antique", "Fable", "Mythos", "Google Piper"], "also_reported_by": [], "alternates": {"html": "https://wpnews.pro/news/scion-how-anthropic-replaced-git-with-centralized-source-control-for-30000-agent", "markdown": "https://wpnews.pro/news/scion-how-anthropic-replaced-git-with-centralized-source-control-for-30000-agent.md", "text": "https://wpnews.pro/news/scion-how-anthropic-replaced-git-with-centralized-source-control-for-30000-agent.txt", "jsonld": "https://wpnews.pro/news/scion-how-anthropic-replaced-git-with-centralized-source-control-for-30000-agent.jsonld"}}