Anthropic built a custom source-control system; small teams probably shouldn’t switch

If your repository history is far below ~100 GB, your daily merged changes are well under 30,000, and your CI does far fewer than 25 million checkouts a day, this change does not save you time or money.

What actually changed

Anthropic replaced standard Git workflows with an in‑house source control system they call "Scion" and a new code review system named Antique. The trigger was a recent surge in engineering velocity: in May 2026 they averaged about 2,500 merged changes per day and by September they were merging over 30,000 per day. Their main branch history is nearly 100 GB and a compressed working copy exceeds 25 GB. Their continuous integration infrastructure performs on the order of 25 million code checkouts per day, not counting developer workstations, agent sandboxes and other environments.

Because those numbers outpaced existing tools, Anthropic rebuilt core systems: a centralised source control model (again, their description sounds similar to older centralised systems such as Subversion) and a bespoke code review flow.

Who it affects

Directly: Anthropic’s internal engineering and CI workloads. The redesign responds to the organisation’s specific scale: very large repo history, tens of thousands of merges per day, and tens of millions of automated checkouts.

Indirectly: only organisations that match those metrics — large model labs, cloud-native companies running massive agent fleets, or firms with similarly heavy CI churn. Small teams, freelancers and most agencies with routine CI and repo sizes far below the figures above will not see practical benefit from replicating Anthropic’s approach.

What it costs or what it replaces.

The announcement describes an internal replacement for Git-based workflows; it replaces Anthropic’s prior Git infrastructure for their needs. The announcement does not state pricing. It also does not say whether Anthropic will release the software or offer it as a product.

Because Anthropic built the systems in-house, adopting the same approach externally would mean committing engineering time to design, implement and operate a custom source-control and review stack. The source indicates those changes were driven by specific scale constraints (repo history near 100 GB, compressed working copy over 25 GB, merge rates growing from ~2,500 to >30,000 per day and CI checkouts on the order of 25 million per day), so the practical cost for most teams is likely higher than improving existing Git workflows or CI configuration.

What we don't know

• Whether Anthropic will open‑source or commercialise "Scion" or Antique.
• Detailed performance trade-offs between Scion and Git for moderate‑scale repositories.
• Ongoing maintenance and operational costs for running Scion at scale.
• Which specific Git-hosting or CI features failed and whether those could be addressed by configuration changes or third‑party services.

What to do next

1. Measure: record your main branch history size, compressed working copy size, average merged changes per day and CI checkout volume. Compare these to Anthropic’s reported thresholds (near 100 GB history, >25 GB compressed working copy, ~30,000 merges/day, ~25 million CI checkouts/day).
2. Optimise before replacing: if you are far below those thresholds, focus engineering time on reducing CI checkouts, caching, and repository hygiene rather than building a custom system. If you approach those thresholds, open a conversation with your Git host and CI vendor about scaling limits and enterprise options.
3. Resource plan: if your measurements match Anthropic’s scale, budget for a sustained engineering project to build and operate custom source-control and review systems rather than expecting a drop‑in product; the announcement shows Anthropic chose an in‑house build for this reason.