Ex-GitHub CEO Thomas Dohmke launches Entire for AI agents

What launched, and when

Thomas Dohmke, GitHub's former chief executive, has launched Entire, a code-hosting platform built specifically to work with AI coding agents, according to CNBC TV18's report on the launch. The company unveiled a "Git forge" on October 26, described as a service for hosting code, reviewing changes and running builds, aimed at teams whose AI agents write code faster than people can review it.

The timing is pointed. The announcement came two days before GitHub's annual Universe conference in San Francisco, according to the report, which cites GeekWire. A former GitHub CEO launching a competitor to GitHub, days before GitHub's own showcase, is a deliberate piece of positioning.

A second account describes the launch as an "agent-first" platform where AI agents work as collaborators with dedicated workspace allocation and access to the same version control primitives as human developers. That description is a secondary summary rather than a company document.

Trails replace pull requests

The core design change is what happens instead of a pull request, the mechanism GitHub uses for proposing and reviewing code. Entire replaces it with Trails, which Dohmke described in a blog post as "customizable evals that are scored by agents on every code push," according to the CNBC TV18 report.

On the platform, multiple AI agents check the quality of every change, so developers can merge code without reading every line. A person's sign-off or automated tests can be enabled before any code goes in. That is the trade the product asks teams to make: review effort moves from reading diffs to configuring evaluations and deciding which checks must pass.

Whether that trade works depends on how good the evaluations are, and neither source reviewed here reports evaluation accuracy, retention data or customer outcomes. The mechanism is a design claim; the evidence that it catches what a human reviewer would catch is not in the launch material.

Hosting, memory and the token claim

Entire is letting developers host private code repositories directly on its service in Europe, the US, India and Australia, according to CNBC TV18. Regional hosting is a practical selling point for companies with data-residency requirements, though the report does not say whether the regions are live for all customers or rolling out.

The company also promises "continuous memory," which allows AI agents to search past coding sessions for the reasoning behind earlier changes. Dohmke said in a LinkedIn post that parts of the feature, Entire Graph and Entire Brain, use up to 71% fewer tokens. The figure is his claim, made in a social post, and the sources here do not describe how it was measured or against what baseline.

Why the launch lands now

The pitch rests on a problem teams already feel. GitHub has been struggling under the load from AI agents and has faced several outages over the past few months, according to CNBC TV18. Code hosting built for human-paced contributions is being asked to absorb machine-paced ones.

Dohmke is not alone in rethinking this layer. Cursor, an AI coding company, launched its own code-hosting service in August, the report says. Microsoft's venture arm M12 was among the institutions that backed Entire's $60 million seed round in February. The category is being contested by incumbents, AI-native startups and the toolmakers whose agents generate the code.

The secondary account adds that the platform is in limited availability with a waitlist and that general availability is planned for early 2027, with pricing based on agent usage rather than seats. Those details are not in the CNBC TV18 report reviewed here, so they should be treated as reported plans from a secondary source rather than confirmed company statements.

What is not yet known

Several important questions are unanswered. Neither source gives pricing for the hosted service, a firm general-availability date, or the names of customers beyond a general reference to early access partners in fintech and cloud infrastructure in the secondary account. Neither reports how many repositories or users the platform has, or what enterprise controls are available beyond agent permissions.

There is also the question of lock-in. A platform that hosts code, runs builds and holds agent memory is infrastructure, and moving it later is expensive. The regional hosting options address one objection, data residency, but they do not answer questions about export formats, migration paths or what happens to the agent memory if a customer leaves.

Finally, the competitive frame is easy to overstate. A launch does not mean GitHub has lost its position. GitHub's scale, with more than 100 million developers according to the secondary account, is not something a new platform reverses in a quarter. The relevant test is whether Entire's agent-scored reviews save review time without letting defects through, and that will only show up in use.

A launch at this stage also says something about the market: code hosting is being rebuilt around agent-generated volume before the old model has visibly broken, which is why several companies are racing to define the replacement rather than waiting for GitHub to do it.

The launch also puts pressure on how teams define review quality, because a Trail that passes is only as good as the evals its operators configure.

How to read the launch

Three claims are worth separating. The product exists and launched on October 26 with hosting, code review and build features. The design replaces pull requests with agent-scored evaluations, which is a real architectural choice with real trade-offs. And the company says its memory features cut token use by up to 71%, which is an unverified vendor claim.

For developers, the practical question is not whether AI agents should be part of code review, since they already are. It is who is accountable when an agent-scored Trail passes and the code breaks. Entire's answer is that a person's sign-off or automated tests can be switched on, which keeps a human in the loop if a team wants one. That is a configuration, not a guarantee.

The launch is best read as a sign of where the developer-tools market is going: the unit of work is shifting from the pull request written by a person to the change produced by a fleet of agents. Platforms built for the first are now being rebuilt for the second, and a former GitHub CEO is betting his next company on the transition. Related coverage includes agent context in Notion 3.8, OpenAI Dots, and Products. For GitHub's own model changes, see the site's report on Copilot model deprecations.