Share with your CTO
Thomas Dohmke, who ran GitHub until late 2024, is betting that centralized Git infrastructure is the wrong foundation for an AI-native development stack. His new company, Entire, launched a preview of a distributed Git network designed for AI coding agents, mirroring repositories across geographic regions to reduce latency and eliminate rate limits. The company raised $60 million at a $300 million valuation at seed stage, has 40 employees across nine countries, and is reporting benchmarks of 570,000 repository clones per hour from a single repo in internal tests.
What this means for your business
The rate-limit problem is already here. Engineering teams deploying even a modest fleet of AI coding agents, say 20 to 50 concurrent agents running on a shared codebase, are hitting GitHub’s API ceilings in ways that interrupt workflows and force manual intervention. Entire is positioning itself as the infrastructure layer that absorbs that read traffic before it ever reaches GitHub’s servers, which is a narrow but genuinely painful problem if you’re scaling agent-based development.
The more interesting product isn’t the distributed mirror, it’s what Entire calls its semantic memory layer. Traditional version control records what changed and who changed it. Entire records the AI conversation, the prompt, and the agent’s reasoning chain alongside the commit. This matters because AI-generated code is increasingly opaque: a developer inheriting a codebase in two years won’t be able to reverse-engineer why a function was written a certain way by reading the diff. Entire Blame, which traces a line of code back to the specific prompt that produced it, could become the audit trail that enterprise compliance teams eventually require for AI-generated software.
The signal worth watching is whether Entire’s open-source commitment holds. The company says it will publish both its Git backend and its benchmarking tools for independent validation. If that happens on schedule, enterprise procurement teams get a credible alternative to simply trusting vendor-published numbers. If it slips, that tells you something about whether the $300 million seed valuation is chasing a real infrastructure gap or a pitch deck narrative. I’d revise this view toward more skepticism if GitHub responds with native regional mirroring, which is well within Microsoft’s infrastructure capability.
Concept deep-dive: Distributed version control vs. centralized hosting
Git was designed as a distributed system, meaning every developer’s local copy is a full repository clone, with no single server required. Commercial platforms like GitHub then built centralized hosting on top of that architecture because discoverability and collaboration needed a hub. AI agents collapse that tradeoff: they clone repositories programmatically, at high frequency, from many locations simultaneously, which turns GitHub’s centralized hub into a bottleneck. Entire is reverting to Git’s original intent, adding geographic mirrors so agents pull from the nearest copy rather than queuing against a single origin server.
Based on reporting from Former GitHub CEO Unveils Distributed Git Network Built for AI Coding Agents, originally published 2026-07-08 13:41:00.
