Redesigning source control at Anthropic
- ID
- 35144
- Status
- summarized
- Published
- 13 Oct 2026, 11:49 PM
- Fetched
- 14 Oct 2026, 9:20 AM
- Provider
- Hacker News
- Category
- dev-community
- Original URL
- https://www.anthropic.com/engineering/redesigning-source-control
- Source URL
- https://hnrss.org/best
Summary
- Score
- 7.0
- Created
- 14 Oct 2026, 9:21 AM
- Tags
- Audience
- developersai_agent_users
What happened
Anthropic reports that its monorepo merges grew from roughly 2,500 code changes per day in May 2026 to over 30,000 per day by September 2026, making git-based source control its biggest development bottleneck. To fix it, the company built Scion, an in-house version control system, and fully migrated its monorepo off git in five months, alongside a code review surface called Antique, code search, and a source control IAM system. It cites main-branch history of almost 100 GB, a compressed working copy over 25 GB, and on the order of 25 million code checkouts per day for CI as the pressures that broke the git model.
Why it matters
The concrete numbers here are a preview of a cost curve, not a tool you can adopt — the post does not say Scion, Antique, or any of the other components are released or open-sourced, so there is nothing to install. What you can act on is the arithmetic: 25 million CI checkouts per day against a 25 GB working copy is the kind of load that turns repository size and CI checkout volume into a real infrastructure bill once coding agents start opening changes at volume. If you run agents that author commits, measure your daily checkout count and repo size now, and decide whether full-clone CI or a monorepo is the right shape before the merge rate — not the code — becomes your bottleneck. Anthropic itself notes its workload is unusual and it does not expect most organisations to hit this in the near term, so treat it as a directional signal rather than an urgent migration trigger.
Discussion angle
Ask the room for their own numbers: how many merges per day and how many CI checkouts does your repo actually do, and at what point would a 100 GB history or a 25 GB working copy stop being workable for your team — and is Anthropic's five-month git-to-in-house-VCS migration a credible engineering trade-off or a sign that AI-generated code volume is outpacing review capacity?