Entire’s new Git forge asks code review to keep up with AI agents

The bottleneck moves downstream

Producing a change and deciding to trust it are different jobs. If AI agents make the first job easier, a development team can end up with more material to inspect rather than less work overall. That is the product problem Entire is addressing with its newly introduced Git forge for collaboration between agents and people.

The company’s October 26 launch post, written by co-founder and CEO Thomas Dohmke, describes three building blocks: Trails for reviewing changes, distributed Git hosting and Continuous Memory for making agent sessions and prompts available with repository context. GeekWire separately reports the introduction of the platform. This is a product launch, not a new funding announcement.

Entire says the service is available to developers now. That statement establishes the company’s advertised availability, not a claim that Foundra has used the platform or verified its performance. The useful examination begins with what information the product proposes to carry across the development workflow.

Review needs context, but not limitless context

Trails is presented as an alternative to relying on a pull-request timeline and a diff. The launch post describes customisable automated evaluations scored on code pushes, with a display intended to help people assess the work. That design points toward a real question: can the reviewer identify which checks matter for the change, rather than merely seeing that checks have occurred?

A hypothetical reviewer might need to understand why an agent changed a dependency, which tests were run and whether the result altered behaviour outside the requested scope. An additional conversation transcript could help answer those questions, but it could also be long, incomplete or misleading. More context is valuable only when the system helps identify the context that bears on the decision.

Automated evaluation also needs a visible boundary. A passing score could establish that a defined check passed; it would not establish that an application is correct in every respect. The platform’s review proposition should therefore be assessed on how it exposes the scope and limitations of checks, not on whether it replaces judgement with a reassuring number.

Memory becomes a repository concern

Entire’s Continuous Memory is described as synchronising sessions and prompts into the repository so agents can search and use that context. The attraction is continuity: a later agent or developer would not have to start from an empty account of the earlier work. The complication is that development conversations may include material whose retention or visibility needs deliberate control.

That is a general governance consideration, not a finding that Entire exposes secrets or mishandles customer data. A team evaluating the product should ask what is retained, who can access it, how sensitive material is handled and how context is removed or corrected. Persistent context changes the information surface of a development system even when the source-code permissions stay familiar.

Distributed hosting raises another set of questions. The company says repositories can be hosted and replicated across multiple regions. A buyer would need to understand the available configurations and their operational implications before turning a geographic capability into a claim about compliance or resilience.

The benchmark is not the buying decision

Entire makes large push-speed comparisons with other services in its launch post. Those are company claims, not independently reproduced benchmarks in this article. The underlying test conditions and a prospective customer’s repository behaviour would matter to interpreting the figures. A faster push does not by itself establish faster delivery of a safe change.

The announcement also names Buildkite and Depot as initial integration partners for continuous integration. That is relevant because source hosting, builds and tests sit in the same delivery chain. Integration availability would still need to be assessed against the buyer’s actual workflow; a partner name is not a full account of how an organisation would migrate.

The decision at the end of the trail

Entire’s proposition is most interesting when viewed as a change in the unit of collaboration. The platform wants the code change, the agent’s context and the evaluation trail to travel together. Whether that helps a team will depend on the quality of the decision it enables at the end: understanding a change well enough to accept it, revise it or reject it.