GitHub's Ex-CEO Raises $60M for AI Agent Version Control That Nobody Asked For
Thomas Dohmke just raised $60 million at a $300 million valuation for a problem that might not exist.
The former GitHub CEO launched Entire last week — a platform that's basically Git for AI agents. While developers are still figuring out if AI coding assistants are actually helpful, Dohmke is already building infrastructure for a future where machines write most of our code.
The Blackbox Flight Recorder Pitch
Entire positions itself as a "blackbox flight recorder" for AI code generation. Instead of tracking human commits, it logs AI prompts, token usage, third-party tools, and troubleshooting guidance. Their first open-source tool Checkpoints integrates with Claude Code and Google Gemini CLI, capturing these AI sessions within existing Git workflows.
<> "We're reimagining the software lifecycle like automotive companies replaced craft-based production with the moving assembly line," Dohmke explains, positioning humans in oversight roles where "trust, oversight, and design strategy" matter more than manual coding./>
Sounds impressive. But here's what's actually happening:
- Developers get searchable indexes of AI sessions tied to commits
- Teams can reuse remediation prompts to prevent repeated AI errors
- Pre-checks avoid duplicate inference calls (saving on API costs)
- Audit trails show which prompts generated which code
The Real Story: Market Timing vs. Market Need
Dohmke's timing feels off. The Hacker News reaction tells the whole story — 338 comments on a 383-point post, with top commenters dismissing the agent hype entirely. One developer put it bluntly: "a single LLM with bash access often outperforms many of the 'agentic systems' they try to sell us on."
That's the brutal truth Entire's $300M valuation ignores.
Most developers aren't drowning in AI agent complexity. They're using GitHub Copilot for autocomplete and maybe ChatGPT for debugging. The idea that teams need specialized infrastructure to manage "hundreds of changes daily" from AI agents assumes a level of AI adoption that simply doesn't exist yet.
Where Dohmke Gets It Right (And Wrong)
Dohmke understands something important: if AI agents do become central to development, we'll need new tooling. Traditional version control wasn't designed for prompt-driven workflows or semantic project queries.
Entire's upcoming semantic layer could genuinely solve problems — letting AI agents query project history directly instead of fumbling through documentation. The ability to track which AI tools generated which code changes addresses real governance concerns for teams already using multiple AI services.
But building for a future that might not arrive is expensive. GitHub succeeded because it solved immediate pain points — centralized repos, pull request workflows, issue tracking. Developers adopted it because they needed it right now, not because it prepared them for some theoretical future.
The Infrastructure Play Nobody Wants Yet
The $60 million seed round signals VCs believe in the "agentic AI" future. Fair enough. But developer tools succeed when they solve today's problems while preparing for tomorrow's.
Entire feels like tomorrow's solution to today's imaginary problem.
Dohmke's 15-person team plans to double soon, racing to build infrastructure for workflows that most developers haven't adopted. Meanwhile, the actual developer experience problems — flaky AI suggestions, context switching between tools, inconsistent code quality — remain largely unsolved.
The real question isn't whether we need better AI development tools. We absolutely do. The question is whether we need these tools for this workflow at this scale.
Based on early developer sentiment, Entire might be building the world's most sophisticated solution to a problem that won't exist for another five years. By then, someone else will probably build it better.
