Seven Anthropic engineers use Claude to build Scion, an in-house version control system, and move the monorepo off git in five months
Daily merges to Anthropic's monorepo went from about 2,500 in May 2026 to over 30,000 in September, mostly because of Claude, and git became the bottleneck. A team of seven built Scion, a new version control system, plus a code review tool called Antique and several related systems, largely with Claude writing the code, and fully migrated off git in five months. Engineers say their main job was steering: telling Claude "no" when it broke core system constraints, and stopping adversarial review from pushing it into over-engineering. They still review every change for intent rather than reading every line.
The numbers (as reported)
| Team | 7 people |
|---|---|
| Time | 5 months to build and fully migrate |
| Merge volume | About 2,500 merges a day in May 2026 to over 30,000 a day in September |
| Infrastructure | Footprint "200x smaller" than the old git mirroring setup; CI got a 25% CPU capacity win |
Facts
- Company / subject
- Anthropic
- Tools
- Claude
- In-house tools
- Scion (Anthropic), Antique (Anthropic)
- Codebase
- greenfield
- Work type
- New product
- Stack
- —
- Published
- 2026-10-13
- Last checked
- 2026-10-23
What the source claims Vendor-published
Claimed by the vendor that sells the tool. Not checked independently.
- New version control infrastructure 200x smaller than the old git mirroring setup in CPU and memory 200x
“Our Scion infrastructure footprint is now 200x smaller than our previous git mirroring infrastructure in terms of CPU and memory requirements.”
- CI and research jobs check out code in seconds rather than minutes, and CI got a 25% CPU capacity win 25% CPU
“CI alone realized a 25% CPU capacity win”
- One rewrite by Claude used 20x less memory and ran 4x faster 20x less memory; 4x faster
“it produced a version that consumed 20x less memory and was 4x faster”
Caveats
Anthropic writing about its own model, so a vendor source. The post doesn't say how much of the code Claude wrote, give a line count or token cost, and notes Anthropic spends far more on tokens internally than other companies would. The infrastructure numbers describe Scion's performance, not the agent's work.
More stories by tool: Claude
Related topics: New products built with agentsGreenfield projectsVendor case studies