Entire proposes trails instead of pull requests for code review with AI agents

Entire, a startup led by Thomas Dohmke (former GitHub CEO), launched a Git forge built from scratch for a scenario where AI agents generate, review, and iterate on code at high speed. The company's bet, according to an InfoQ report, is that the pull request model that has sustained GitHub and GitLab for more than a decade was not designed for this pace.

Dohmke's argument is direct: GitHub and GitLab organized human-to-human collaboration around asynchronous code review and versioning. A multi-agent forge, in his view, needs to organize around something else: the orchestration, versioning, and asynchronous auditing of AI agents working alongside humans.

The problem Entire says it has identified

The company's central criticism of the traditional PR is that the timeline of comments and commits no longer serves the human developer. In practice, it has become a record that only the AI agent itself consults to know what it did.

The pull request [timeline] has become an audit log that only the agent reads. It does not provide the information for the developer to decide whether they can merge with confidence.

Thomas Dohmke, CEO of Entire

Instead of scrolling through dozens of comments and incremental commits from an agent working alone, Entire's proposal concentrates at the top of the page the information that, according to the company, really matters for deciding the merge.

Trails: eval score on every push

The product's central component is called trails. A trail can originate from a bug report, a feature request, or be opened manually, and it can span multiple repositories at the same time.

On every push of new code, a review agent runs in a cloud sandbox and returns a score with justification, covering four axes: code quality, security, "slop" (code generated carelessly or redundantly), and need for human review. If the slop score comes back orange, for example, the developer can ask the coding agent to improve the snippet and push a new change, opening another iteration of the cycle.

The interface shows the last five evaluation runs plus the current run side by side, and this is described as a deliberate design decision:

The last 5 runs and the current run are shown for each eval. We believe this is crucial as one-shot loops are often missing a critical issue. Repeating the loops and stabilizing the scores is crucial for the success of agentic coding.

Thomas Dohmke, CEO of Entire

The evaluation agents can be configured through Entire's interface or via a JSON file at the repository level, giving teams control over which criteria carry more weight in each project. Teams can also define Gates at the repository level, specifying mandatory approval conditions. While the trail is open, every finding flagged by the reviewing agent must be resolved or closed before merging: the merge only happens when confidence is high, risk is low, and no pending findings remain.

Continuous memory: the context that disappears between sessions

The product's second pillar is continuous memory, designed to solve a specific problem with agents: every new session starts from scratch, without access to the reasoning that led to previous code decisions. Entire implements this with two open source tools:

• entire-graph: builds a pre-computed map of the Git repository, with ranked code search, mapping of definitions and function calls, types, and routes;
• entire-brain: creates a persistent memory layer combining this code map with previous agent sessions and the history recorded in Git.

The idea, according to the company, is for the agent to always have the right context at the right time: the "why" behind the "what changed," not just the raw diff. Entire also announces integration with CI platforms such as Buildkite and Depot.

The reaction on Hacker News

The launch sparked debate on Hacker News. A user identified as conartist6 argued that engineers started delegating code writing to agents because programming languages and tools leave much to be desired. Haimantika, who now works at Entire, disagreed with that diagnosis:

it's that the reasoning gets thrown away between the person and the artifact. Commit messages were always bad, docs always rotted, "why is this here" was answered by whoever remembered. Agents just made that acute, because now there's no one who remembers. That's the bit we're trying to fix. If a better language fixes it too, honestly I'd take that.

Haimantika, engineer at Entire

Entire is not alone

InfoQ lists Entire as one of several new players competing in the source code management space geared toward agentic workflows: Zed DeltaDB, Anthropic Scion, East River Source Control, Cursor Origin, and Radicle also appear in the same category. It's a sign that GitHub and GitLab (which have in parallel been beefing up agentic features, as the launch of GitLab 19.5 showed) are no longer the only possible addresses for whoever decides where to version AI-generated code.

What this changes for whoever reviews code here

For Brazilian teams that have already put Claude Code, Cursor, or Copilot into their workflow, Entire's message is less about switching forges tomorrow and more about naming a real problem: a giant PR, generated by an agent, reviewed by a tired human, is review theater, not review. The per-push eval score and the configurable Gates attack this point directly, bringing an objective criterion (even if imperfect) before asking for human attention.

The most replicable part, regardless of switching platforms, is the concept of continuous memory: teams that currently rely on repeated prompts for every agent session can look at entire-graph and entire-brain, both open source, as an architectural reference, not just a proprietary feature. A healthy dose of skepticism is also warranted: this is a new product, from a newly founded company, competing against GitHub and GitLab with years of CI/CD pipeline integration, corporate SSO, and compliance already worked out. Before migrating a production repository, the more sensible path is to test the trail on a small project and compare merge time against the current PR workflow.