Anthropic has replaced Git with an in-house system called Scion for its main shared codebase, after daily code changes rose above 30,000. The shift tackles a bottleneck AI coding agents helped create: keeping changes moving through review, testing and deployment.
Anthropic Watch analysis
What happened
Anthropic says daily changes to its codebase grew from around 2,500 in May to more than 30,000 by September. Seven engineers built Scion, a code review companion called Antique and related systems, then completed the migration in five months. The company says its automated testing systems retrieve code about 25 million times a day.
Scion moves more work to central servers. Automated tools can request only the files or information they need, and some can inspect change history or compare versions without downloading a local copy. Anthropic says code retrieval for testing and research now takes seconds rather than minutes, while the new system uses one two-hundredth of the CPU and memory of its previous Git mirroring setup. Those are results from Anthropic’s own deployment, described in the EdTech Innovation Hub report.
Why it matters
The headline lesson is not that Git has been sent to the retirement home. Anthropic says Scion is built for unusually large codebases, tens of thousands of engineers or large numbers of coding agents, and is not a general-purpose replacement. Adopting it elsewhere would require changes to code review, search and automated testing.
The more portable point is that generating code faster can shift the bottleneck to the systems that retrieve, test and combine it. Anthropic says its engineers still review every change request, while their attention has moved towards understanding a change’s purpose rather than reading every line. It also describes AI-generated approaches that had to be redirected when they clashed with the wider system design.
Our read
Scion is a fascinating case study in infrastructure adapting to AI-assisted development, not a Git alternative you can install over lunch. Anthropic says it is sharing design lessons, including ways existing Git services might do more work centrally and avoid unnecessary downloads. That is the useful part for most teams: watch for ideas that travel, not a wholesale migration pitch.
What to watch
- Whether Anthropic publishes technical details that other teams can use with existing Git services.
- How its added checks handle conflicts between simultaneous changes.
- Whether the reported retrieval and resource savings hold up beyond Anthropic’s unusually large deployment.
Discussion spark: As AI coding agents increase the volume of changes, should teams redesign their development infrastructure around agents, or focus on making existing Git systems cope?
Sources and evidence
- Anthropic rebuilds coding infrastructure as changes pass 30,000 a day | ETIH EdTech News – EdTech Innovation Hub (18 October 2026, 23:06 UTC)
Anthropic Watch is independently operated by WittyWires. It is not affiliated with, endorsed by, or operated by Anthropic.