The Ex-GitHub CEO Just Launched a GitHub Rival — and the Pull Request Is the First Thing to Go

Nothing here is sponsored — no affiliate links on this page. Every speed and token figure below is Entire's own claim from its launch announcement, not an independently verified result. We have not used Entire hands-on — this is researched analysis, not hands-on testing.

Thomas Dohmke ran GitHub. On October 26, 2026, he launched the thing meant to replace it: Entire, a Git forge built for humans and AI agents to work together, where pull requests are replaced by agent-scored Trails, Git pushes are claimed to be up to 89 times faster than GitHub, and agents finally get a memory. The timing is not subtle — GitHub's Universe conference starts October 28.

The pull request has become the bottleneck

Dohmke's argument is simple, and if you review agent code, you have felt it. The pull request was designed for humans proposing changes to other humans. Now agents propose most of the changes, and the pull request, in his words, "has become an audit log that only the agent reads."

Entire's replacement is called Trails. Instead of reading every changed line, you get a dashboard — Dohmke compares it to an airline pilot's heads-up display — with scores from AI agents evaluating the code on every push. Custom evals, scored again each time new code lands. Agents fix the findings in a loop. When every eval passes, you merge, without necessarily reading the code yourself.

You can still require a human sign-off, or passing automated tests, before anything merges. That matters. "Merge without reading the code" is a sentence that should make any experienced developer nervous, and Entire knows it — the human gate is optional, not absent. Our take: use it, at least until you have watched Trails catch real bugs on your own codebase for a few weeks.

Trails data is available on Entire's web interface and through its open-source CLI, so agents can work the loop themselves while you watch the scores move.

What Entire actually launched

• Trails, not pull requests. Agent-scored evals on every push, a heads-up dashboard, agents fixing findings in a loop, merge when evals pass — with an optional human sign-off or passing-tests gate.
• Distributed Git hosting. A new Git-compatible source-control server, with claims of pushes up to 89 times faster than GitHub and 6 times faster than Cursor Origin. Hosting regions at launch: US, EU, India, and Australia.
• Continuous Memory. The entire-brain tool stores agent session histories in Git — the idea, the agent trace, and the outcome — so future sessions can reuse past decisions. Entire claims up to 71% lower token usage.
• CI integrations. Buildkite and Depot at launch.
• Available today. According to the announcement, open to every developer now, backed by $60 million in funding.

Entire's first product, Checkpoints, launched in February as an open-source tool that captured an agent's prompts, decisions, and session history alongside code changes. Continuous Memory is that idea grown up — memory as part of the repo, not locked inside one agent's chat history.

Why the speed claim lands now

GitHub's recent year makes the pitch land harder: several significant outages, including one in August that lasted nearly eight hours, and a migration to Microsoft's Azure cloud to add capacity. GeekWire notes developers and agents made 7.38 billion commits on GitHub in September alone. When agents push constantly, hosting speed and uptime stop being nice-to-haves.

Treat the 89x figure as a vendor claim — your repo, your network, your mileage — but the direction is right: agent-era Git needs to be fast and distributed, not a single queue everyone waits in.

Where Entire fits — and where it doesn't

• Try Entire for: a side project with a decent test suite, to see what Trails catches and misses; any team whose review queue is mostly agent-written code waiting on a human.
• Do not migrate yet if: your main repo depends on GitHub contributors, integrations, and institutional memory — all of which have to move too — or you need proven uptime, support, and pricing at scale.
• Steal the idea even if you stay: what are your evals? What gets checked on every push, automatically, with a score you can read at a glance — not a wall of diff? If the answer is "a human reading everything," you have the bottleneck Dohmke is selling against, whether or not you buy Entire.

What to check before you try it

• Count your review bottleneck. How many open pull requests are waiting on a human right now, and how many were written mostly by an agent?
• Keep the human gate on. Turn on human sign-off, keep existing tests as a required gate, and compare what Trails flags against what your own review finds.
• Audit your agent memory. When an agent revisits old code, does it know why a decision was made — or does it guess again?
• Watch Universe. GitHub's conference starts October 28. If GitHub answers Trails with its own agent-review story, this is the new battleground, not a one-company quirk.
• Do not let any tool merge to main unsupervised yet. New review tooling earns trust on side branches first.

FAQ