Machines Are the Primary Producers Now

There is a specific kind of founder we spend our lives looking for: the one who built the thing that created the problem, then went back to zero to fix it.

Thomas Dohmke ran GitHub for four years. He took over in 2021 and presided over the period when Copilot went from a curiosity to the default way a large share of the world’s code gets written. In August 2025 he left. In February 2026 he came out of stealth with Entire, backed by a $60 million seed at a $300 million valuation, described by lead investor Felicis as the largest seed round ever raised by a developer tools company. Madrona, Microsoft’s M12, Basis Set, 20VC, Cherry, Picus and Global Founders Capital came in alongside, as did Jerry Yang, Garry Tan, Datadog’s Olivier Pomel, and several of the most-read voices in the developer community.

The problem Entire exists to solve is the one Dohmke helped create.

The audit trail vanished

Agents now write enormous volumes of code, faster than any human can meaningfully read. The artifact that survives is the diff. Everything that produced it (the prompt, the constraint, the reasoning, the three rejected approaches, the decision about why this schema and not that one) evaporates into a chat window nobody keeps.

Which means the industry has quietly lost something it spent forty years building: the ability to explain why the software is the way it is. Code review assumed a human author who could be asked. Git assumed a commit was a human decision. Pull requests assumed a reviewer with time to read. None of those assumptions hold when the producer is a fleet.

Dohmke’s framing is a manufacturing one, and it is the right one: software is living through the shift from craft production to the assembly line, and the development lifecycle we all use was designed for the craft era.

Entire’s first shipped product, an open-source CLI called Checkpoints, does the unglamorous foundational thing: it captures agent context alongside the commit, so the reasoning behind machine-written code becomes durable engineering data rather than disposable chat. Underneath sits a git-compatible database unifying code, intent, constraints and reasoning in one version-controlled system; a semantic reasoning layer that lets multiple agents work together; and an interface built for agent-to-human collaboration rather than retrofitted for it.

Notably, Entire is not building a model or a coding agent. It integrates with them: Claude Code, Cursor, Codex, whatever wins. That is a deliberate position and a durable one. The layer that governs and remembers is more defensible than the layer that generates, because the generators are in a price war and the governance layer is not.

The forged part

Dohmke is from East Germany. He built a startup, sold it to Microsoft in 2015, moved to the United States, and worked his way to running one of the most important pieces of infrastructure in software. Then he left the top job at a Microsoft subsidiary to go be fifteen people again.

We back that decision profile deliberately. A founder who has already achieved the thing most people spend a career reaching for, and who walks away from it because the interesting problem is elsewhere, has removed the most common failure mode in venture: the founder optimising for status without knowing it. There is nothing left for Dohmke to prove about his résumé. The only thing left to prove is the product.

On the size of the round

We should be honest about an apparent tension. We argue that founders should build companies that are not maximally dependent on the capital markets, and Entire raised one of the largest seeds in its category’s history. The distinction matters more than the headline number.

The thing to avoid is dependence, not capital. A company is dependent when every technical milestone triggers an existential financing judgment, when the only route to the next proof point is a larger cheque from someone who now controls whether you continue. Entire is not shaped that way. It ships open-source tooling developers adopt without a procurement cycle. It sits alongside coding agents enterprises already pay for. Its cost structure is a small team and, increasingly, tokens. Dohmke has been blunt that in 2026 any leader planning headcount has to think about token spend the way they think about salaries.

That is a company that raised a lot because the opportunity is large and the founder could, not because the physics demanded a decade of pre-revenue capital. Different animals wearing the same number.

Why we’re here

Every generation of software gets a system of record, and the system of record ends up worth more than most of the applications sitting on it. Source control was that layer for the human era. Nobody has built it yet for the era where machines produce the code and humans supervise the fleet.

We think that layer gets built once. We think it gets built by someone who ran the last one.