Entire

What Entire is building

Here's the problem Entire is solving in one sentence: AI can write code faster than humans can understand it. GitHub Copilot, Cursor, and Claude are generating production code at a pace that has broken the traditional human-review model. A PR might contain 2,000 lines of AI-generated code, and the reviewer has no idea what prompted it, what constraints the AI was working within, or what alternatives it considered. Entire's first product, Checkpoints, solves this by creating a new kind of repository: a Git-compatible database that stores not just the code, but the full creation context - the prompts, the reasoning sessions, the decision trees. Their semantic reasoning layer then lets humans and AI agents query across this context, not just the code. This repository structure enables auditing the 'why' behind changes, not just the 'what,' by capturing atomic steps, configuration, and state updates, advancing beyond traditional code-only storage, such as Github. The phrase Thomas Dohmke uses: 'We're moving away from engineering as a craft, where you build code manually and in files and folders [...] to a much higher abstraction, which is specifications - reasoning, session logs, intent, outcomes.' Version control for the age of agentic software.

Why this matters

The numbers make this concrete. GitHub Copilot users report that AI now writes between 30-40% of their code. At companies running multiple AI agents in parallel - the direction every major engineering platform is moving - that figure is higher. A senior engineer reviewing a pull request in 2026 may be looking at 2,000 lines of code they didn't write, can't fully trace, and are being asked to approve in a sprint cycle measured in days. The context that produced that code - the prompt that initiated it, the constraints the agent was working within, the alternatives it considered and rejected - exists nowhere. It evaporated the moment the session ended. Git, the version control system that underpins virtually all software development, was designed for a world where humans wrote every line and could explain every decision. It records what changed. It has no architecture for recording why, and no concept of an AI agent as an author with its own reasoning chain. The entire code review and accountability infrastructure of modern software engineering is built on an assumption that code has a human author who can be questioned. That is becoming less true every month. The deepest impact is on multi-agent software development. The industry is moving rapidly toward architectures where multiple AI agents collaborate on a single codebase simultaneously - one agent writing tests, another refactoring, another generating documentation, all operating in parallel. The coordination problem this creates is orders of magnitude more complex than human team coordination, because agents have no shared memory, no implicit understanding of each other's reasoning, and no way to resolve conflicts without a system that explicitly stores and shares context. The market this opens is significant. Developer tools is a $26B+ market growing at 20%+ annually, and the tooling layer for AI-generated code is almost entirely unbuilt. Every company running AI coding agents - which will be effectively every software company within three years - will need some version of what Entire is building. The $60M seed, the largest in developer tools history, reflects how much conviction Felicis and M12 (Microsoft's venture arm) have that this is a category-defining bet. The Microsoft angle is particularly notable: Satya Nadella told Dohmke personally to 'keep pushing until your last day' and supported his departure amicably. M12 investing in Entire signals Microsoft sees this as complementary infrastructure, not a threat.

Investors: Felicis, Madrona, M12 (Microsoft), Basis Set