Anthropic has replaced Git, the industry-standard version control system, with a custom-built alternative designed to keep pace with the volume of code that Claude now produces. The decision, reported by TechSpot, reflects a growing tension between legacy developer infrastructure and the speed at which AI models can generate, commit, and iterate on code.
Why Git Was No Longer Enough
Git was built for human developers working at human speed. Branches, merges, and pull requests were designed around workflows where a single engineer might push a few commits per day. That model breaks down when an AI system is generating code continuously and at scale. According to Anthropic, Claude is now writing 65% of Anthropic's own code, a figure that puts enormous pressure on any version control system built with human throughput in mind.
Key Facts
- Anthropic has moved away from Git to an internally developed version control system.
- The switch was driven by the volume and velocity of code Claude generates.
- Claude now authors the majority of code inside Anthropic's own codebase.
- The move highlights broader infrastructure challenges facing AI-native development teams.
- This is one of several internal tooling changes Anthropic has made to support Claude-driven workflows.
The problems with Git in this context are practical ones. Merge conflicts multiply when dozens of AI-generated branches exist simultaneously. Commit histories become unwieldy. Review workflows designed for human-authored pull requests do not translate cleanly to AI-generated code, where the volume alone can make meaningful oversight difficult. There is already concern in the developer community that Claude Code is writing pull requests that developers never fully read, and a version control system that compounds that problem is a liability.
The infrastructure we use to build software has to match how software is actually being built today.Anthropic, via TechSpot
A Broader Shift in Developer Tooling
Anthropic's decision is part of a wider pattern of internal tooling adjustments. The company has also been rethinking source control from the ground up for an AI-coded world, a project that now appears to have produced a concrete product for internal use. Separately, the team behind Claude Code has been working to streamline the system itself, including cutting its system prompt by 80 percent to improve efficiency and reduce overhead.
The common thread across these efforts is adaptation. When an AI model becomes a primary contributor to a codebase, every layer of the development stack needs to be reconsidered. Version control is one of the most foundational layers, which makes Anthropic's move significant beyond its immediate practical benefits. Other organizations leaning heavily on AI coding tools will face the same reckoning eventually. Whether they build their own solutions or wait for the open-source ecosystem to catch up remains to be seen.
For now, Anthropic's custom system is internal, and the company has not indicated plans to release it publicly. But the fact that a leading AI lab found Git insufficient for its own operations is a signal the developer tools industry will likely take seriously. The companies building on top of models like Claude's are watching closely, and the pressure to modernize version control for AI-native teams is only going to increase as models grow more capable and more deeply embedded in production codebases.
