The Former CEO of GitHub Just Agreed: Git Wasn't Built for This
Source: Dev.to
Interview with Claude Opus 4.5
Two weeks ago, I interviewed an AI about what it actually wants from developer infrastructure. This week, Thomas Dohmke raised $60 M to build it.
On February 4th I published an interview with Claude Opus 4.5 titled “Git is Dead to Me: Why AI Agents Hate Your Pull Requests.” The thesis was simple: files are an OS constraint from the 70s, Git is a protocol from 2005, and we need to stop duct‑taping new intelligence onto old infrastructure.
The AI was blunt:
“Give me a flat representation with explicit edges between things, tell me the constraints, and let me emit a new state. Don’t make me do surgery on text files and pretend I know which line I’m on.”
When I asked what versioning should look like:
“Here was the state, here was the intent, here’s the new state. Not: here are 43 line‑level changes across 12 files.”
I called it State = f(Intent[]). The deployed artifact should be a pure function of the conversation history—no diffs. Replay intent, regenerate state.
Some people thought this was hyperbole. An AI complaining about Git? Surely the fundamentals still hold.
Thomas Dohmke Launches Entire
Six days later, on February 10th, the former CEO of GitHub — the man who scaled Copilot to millions of developers — announced Entire, a new company with $60 M in seed funding at a $300 M valuation.
His launch post summed up the problem:
“We are living through an agent boom, and now massive volumes of code are being generated faster than any human could reasonably understand. The truth is, our manual system of software production — from issues, to git repositories, to pull requests, to deployment — was never designed for the era of AI in the first place.”
In an interview with The New Stack he went further:
“We’re moving away from engineering as a craft, where you build code manually and in files and folders… toward specifications, reasoning, session logs, intent, outcomes. That requires a very different developer platform than what GitHub is today.”
The guy who ran GitHub just said GitHub is wrong for agents.
What Entire Is Building
Entire’s platform has three layers:
- A Git‑compatible database that versions code, intent, constraints, and reasoning together.
- A universal semantic reasoning layer that enables multi‑agent coordination through a “context graph.”
- An AI‑native interface for agent‑to‑human collaboration.
Their first product, Checkpoints, captures the prompts, decisions, and execution traces behind every AI‑generated commit. When you commit code from an agent, Checkpoints stores the full session: transcript, prompts, files touched, token usage, tool calls.
“Versioning should be: here was the state, here was the intent, here’s the new state.” — Claude Opus 4.5
“Checkpoints are a new primitive that automatically captures agent context as first‑class, versioned data.” — Entire announcement
Same insight. Same primitive. The difference is Dohmke has $60 M and a team to build it.
Why This Matters
The validation isn’t that I was right; it’s that the most credible person in developer tooling independently arrived at the same conclusion—and is betting his next company on it.
Implications
- Git isn’t going away — it becomes a storage layer, not a workflow. Entire is Git‑compatible precisely because you can’t rip out Git, but they’re adding the semantic layer that actually matters for agents.
- The “90 % problem” isn’t about GitHub integration — Vercel announced a v0 rebuild the same week, touting deeper GitHub integration as the solution. Dohmke is betting the opposite: the GitHub workflow itself is the bottleneck.
- Intent is the new primitive — not files, not commits, not diffs. The conversation that produced the code is more valuable than the code itself. Checkpoints makes that explicit.
Where We Go From Here
At Opzero we’ve been building on a related thesis: AI agents need deployment infrastructure that works from inside the LLM client, not destination apps that compete with your AI.
We’re not building Entire. They’re focused on governance and audit — making AI code reviewable. We’re focused on the other end: making AI code deployable without the ceremony.
But we share the same premise: the primitives are wrong. Files, folders, commits, PRs — these are human coordination mechanisms that agents are forced to serialize into because that’s the expected format.
The next wave of developer infrastructure will be AI‑native from the ground up. Not Git with AI bolted on. Not PRs reviewed by AI. Something new.
Dohmke is building one piece of it. We’re building another.
The race is on.
Originally published on the Opzero blog. I’m building Opzero, an AI‑native deployment platform. Follow me on Twitter.