Blog

GitHub’s Former CEO Just Launched Entire. Here’s What It Means for Code Review

Thomas Dohmke’s new Git forge swaps the pull request diff for “Trails,” where review agents score every push. Here’s what Entire launched, how to read its speed claims, and why I’m still keeping a human at the gate.

Caleb Dorris 26 October 2026, 8:08 PM Updated 27 October 2026 8 min

Long-exposure light trails on a curving highway at dusk with a city skyline behind, under the headline Agents Review Every Push.

The short version

Code review is starting to move from the diff to the scorecard.

  • What happened: on October 26, GitHub’s former CEO Thomas Dohmke launched Entire, a Git forge built for AI agents and people working together.
  • The big idea: “Trails” replace the classic pull request diff with customizable evals that agents score on every push. You can still require human approvals and passing CI before a merge.
  • The speed claim is Entire’s: its own open-source benchmark says pushes are up to 89x faster than GitHub and 6x faster than Cursor Origin, measured as many concurrent pushes, the way a fleet of agents hits a server.
  • My take: agents doing the first pass on review is where this is all heading. An eval score still isn’t an approval, so I’m keeping a human at the gate.

Update, October 27, 2026

A second example: SuperPlane 1.0 takes agents from ticket to pull request

The day after Entire launched, Semaphore, the CI/CD company, released SuperPlane 1.0, an open-source "software factory" for coding agents. Entire rethinks review once the code exists. SuperPlane goes after everything around it: picking the work, running the agent and chasing the checks.

What Semaphore says it does:

  • Scores backlog tickets "for clarity, complexity and verifiability" and helps fix the weak spots before an agent runs.
  • Runs a coding agent on each task in its own cloud dev environment and opens a pull request.
  • When CI fails or a reviewer comments, it "loops the agent back until the PR is ready for a person."
  • Hands back evidence with each change: passing checks, screenshots, recordings "and a note on anything it couldn't verify."

Semaphore says it works with models from OpenAI, Anthropic, Google and more, and the docs list integrations for Claude, Cursor, OpenAI and OpenRouter. The core is Apache 2.0 on GitHub, and you can self-host the community version with Docker Compose on one Linux VM. Two bits of fine print: files in the repo's /ee folder fall under a separate Enterprise license, and the Kubernetes install on AWS, Azure or Google Cloud requires an Enterprise license. There's also a hosted version with pay-as-you-go usage pricing.

About that 52%. Semaphore's exact words: "Last month, 52% of the pull requests SuperPlane engineering team merged were automated through the factory." That's their claim about their own team, and I haven't verified it. It also doesn't mean half their PRs merged themselves. SuperPlane’s docs say "a person remains responsible for decisions such as accepting uncertain work, reviewing important changes or merging the final pull request," though depending on how a team sets it up, a verified PR can be "ready for normal team review or already approved for merge."

That's the same line I drew above. Agents can do the work, chase failing checks and score the result. A person still decides what ships. Semaphore suggests starting with "a small, well-defined bug," and that's how I'd try it too: one low-risk repo, a human approval on every merge, and a side-by-side look at what it hands back versus what I'd have written. If you want help setting up agent workflows with a human at the gate, let's talk.

On October 26, Thomas Dohmke, who used to run GitHub, officially launched Entire after teasing it in July. Entire calls itself “the first Git forge built specifically for agent-human collaboration,” and the headline feature goes straight at the pull request.

That caught my attention, because review is the part of agent-assisted work that doesn’t speed up on its own. Agents can write code fast, and now they can even test what they build. Somebody still has to decide what gets merged. Entire’s answer is to hand most of that first pass to more agents.

What Entire actually launched

According to Entire’s launch post, the platform rests on three pieces:

  • Trails. In place of reviewing a diff, a Trail gives you what Entire calls a heads-up display “similar to an airline pilot,” with “automated customizable evals that are scored by agents on every code push.” A Trail can start from a bug report, a feature request or your agent, and it can span more than one repository.
  • Distributed Git hosting. A new Git-compatible server you can replicate across regions. Entire says it’s live in the US, EU, Australia and India, with data residency options.
  • Continuous Memory. Agent sessions and prompts get synced into the Git repository so coding agents can search them later.

Entire also partnered with Buildkite and Depot for CI and says it’s “available for any developer today.”

How Trails change code review

This is the part I think matters most, whether or not Entire wins. Here’s how Entire describes it:

  • Review agents run on their own in isolated cloud sandboxes from Daytona and E2B, using Claude Code and Codex. Each one returns a score and a rationale.
  • There are pre-built agents for code quality, security and “slop” that you turn on per repo. You can define your own in the UI or as .json config in the repo.
  • Findings come with severity levels and suggested fixes. An agent or a person can address them or dismiss them.
  • “Gates” set what has to happen before a merge, including human approvals and passing CI checks.

One nuance worth knowing: if you mirror a GitHub repo, Entire says you can run Trails on top of your existing pull requests. It replaces the diff-first review model. It doesn’t force you to give up PRs.

About that 89x number

The launch post says: “According to our open-source benchmarks, pushes are up to 89x faster than GitHub and 6x faster than Cursor Origin.” Its chart lists pushes per second: Entire 418, Cursor 65, Pierre 15, GitLab 10 and GitHub 4.7.

Those are Entire’s numbers, not mine, and I haven’t run them. The benchmark is ForgeMark, an MIT-licensed tool that measures “sustained push or clone throughput” with many concurrent writers, “the way an agent fleet hits a forge.” So this is about how a server holds up when lots of agents push at once. It doesn’t mean your single git push will feel 89 times faster. ForgeMark’s README also warns you to run it only against infrastructure you own or are authorized to load-test, so don’t aim it at someone else’s server.

Continuous Memory, and what ends up in your repo

Continuous Memory runs through two new Entire CLI commands. entire-graph maps how your code fits together, and Entire says it “saved up to 71% on tokens” in its own testing. entire-brain indexes your agent sessions so an agent can pull up past decisions. Both are open source and local-first. A team can optionally connect entire-brain to Entire’s backend to share memories.

I like the idea of keeping the “why” next to the code. But if prompts and sessions are synced into the repo, anything pasted into a prompt can be synced too. Keep secrets out of agent chats and scope what your agents can touch before you turn on anything that saves sessions.

Where I think code review is headed

Agents taking the first pass is already happening. Cursor says the harness behind its Agent SDK also powers its own Bugbot and Security Review agents. Entire builds that same idea into the forge and makes the score the main thing you look at.

I think that’s mostly good. A human reading every line of every agent push doesn’t scale. But look at what actually stopped the AI agent in that recent GitHub supply-chain test: a curious person, a maintainer who said no and a platform setting that held untrusted code until a human approved it. An eval score is a great filter. It isn’t the same thing as someone accountable saying “ship it.”

So here’s how I see it. Agents score every push. People review the scorecard, the risky files and anything an agent flags or can’t explain. A Gate that requires a human approval stays on.

What I’m doing about it

I’m not moving any repos on launch day. New tools get a trial, not a switch. Entire’s mirror option keeps GitHub as the source of truth, so if I try it, I’d mirror a side project, turn on the security and slop agents, require a human approval Gate, and compare what they catch with what I catch. I didn’t see pricing in the launch post, so I’d check that before relying on it for anything real.

If you’re trying to figure out how AI review should fit into your team’s workflow, your site or your app, let’s talk.

FAQ

What is Entire?
A Git forge for AI agents and people working together, launched October 26, 2026 by Thomas Dohmke, GitHub’s former CEO. It combines Trails (agent-scored review), distributed Git hosting and Continuous Memory.

Does Entire replace pull requests?
It replaces the diff-first review model with Trails. If you mirror a GitHub repo, Entire says you can use Trails on top of your existing pull requests.

Is Entire really 89x faster than GitHub?
That’s Entire’s claim from its own open-source ForgeMark benchmark. It measures pushes per second with many concurrent writers, not how fast a single push feels. I haven’t verified it.

What is Continuous Memory?
Entire’s way of syncing agent sessions and prompts into the repo so agents can search them. It runs through the open-source, local-first entire-graph and entire-brain CLI tools, with optional team sync.

Can I try it without moving off GitHub?
Yes. Entire says you can mirror a GitHub repo and keep GitHub as the source of truth.

Do agent reviews mean I can skip human review?
I wouldn’t. Entire’s own Gates can require human approvals, and I’d keep that switched on.

Sources

Need a web developer in Nashville?

PHP, WordPress and Next.js. Free quote, and I usually reply within a couple of hours.

caleb@calebdorris.com and call or text (615) 914-1073

Get a free quote

Nashville web development What a free quote covers

All posts