Anthropic has reportedly replaced Git internally with Scion, a proprietary source-control system, after AI coding agents helped drive a day of more than 30,000 code merges. The reported switch is a striking sign that agent-heavy software work may be changing not just how code gets written, but the infrastructure used to manage it.
Anthropic Watch analysis
What happened
Hackernoon reports that Anthropic moved to Scion after automated agents contributed to 30,401 code merges in one day. The article describes Scion as a server-centric system, rather than one built around full local repositories, and says it reduces checkout and commit times for Anthropic’s engineers and agents.
The system is proprietary and designed for Anthropic’s internal workload, according to the report. It gives no public performance figures in the supplied account, so the scale of any speed-up remains unspecified.
Why it matters
The interesting part is not simply a company swapping one version-control tool for another. If the account is accurate, a very high volume of agent-assisted changes helped make Git’s familiar model a bottleneck worth replacing. That puts code review, coordination and repository infrastructure in the same conversation as the coding agents themselves.
Scion is not a product teams can adopt from this announcement, and one company’s internal migration does not establish that Git is suddenly past its sell-by date. But it is a useful glimpse of the operational adjustments that may follow when AI agents produce code at unusual scale.
Our read
The reported merge count is eye-catching; the more durable story is that agent use may force changes deeper in the software stack than a new coding interface. Anthropic’s account, as described by Hackernoon, makes Scion sound like a response to its own workload, not a general-purpose replacement developers should rush to seek out.
What to watch
- Whether Anthropic publishes technical details or performance measurements for Scion.
- Whether other companies report source-control bottlenecks from large-scale coding-agent use.
- How Anthropic checks and reviews agent-produced changes at this reported volume.
Discussion spark: If AI agents make code changes faster than existing tools can manage them, should companies rebuild core developer infrastructure, or first put more emphasis on limiting and reviewing agent output?
Sources and evidence
- Claude Merged 30,401 Changes in One Day. Git Couldn't Keep Up. (13 November 2026, 07:56 UTC)
Anthropic Watch is independently operated by WittyWires. It is not affiliated with, endorsed by, or operated by Anthropic.