Skip to content
MAR
Sign inRequest access
Agentic AI

Entire Rethinks Git for AI-Human Code Collaboration

A new Git forge prioritizes asynchronous orchestration and transparency for high-velocity AI-assisted development workflows.

MrBarisCMrBarisC@barisMarsel TeamCreator
CrawledInfoQ

InfoQ

Traditional Git platforms like GitHub and GitLab streamlined human collaboration through asynchronous code review and version control. But when AI coding agents join the mix—generating, reviewing, and iterating on code at machine speed—the pull request model breaks down. Entire, a Git forge founded by former GitHub CEO Thomas Dohmke, is purpose-built to handle this new reality.

Why the PR Model Falls Short for AI#

The core problem isn't technical speed; it's visibility and confidence. In traditional workflows, a PR's commit timeline becomes an audit trail that developers read to decide whether to merge. But when an agent generates dozens of iterations in minutes, that timeline becomes noise—developers can't tell if the code is stable or still evolving. Dohmke frames the shift: a multi-agent forge must "center around the asynchronous orchestration, versioning, and auditing of AI agents working alongside humans."

Trails: A New Review Primitive#

Entire's breakthrough feature is trails, which replace the PR-as-audit-log model with actionable decision support. Each trail carries eval scores that update after every push, showing developers at a glance whether code is ready to merge or needs another agent iteration. These scores assess code quality, security, code style ("slop"), and whether human review is needed.

A trail can originate from a bug report, feature request, or manual trigger, and can span multiple repositories. When code is pushed, a review agent running in a cloud sandbox assigns a confidence score and rationale. If the slop score turns orange, a developer can ask the coding agent to refine and push again—triggering a fresh evaluation cycle.

Crucially, the interface shows the last five runs plus the current one. Entire argues this history matters: "one-shot loops are often missing a critical issue. Repeating the loops and stabilizing the scores is crucial for the success of agentic coding."

Continuous Memory: Context Without Amnesia#

A second innovation is continuous memory—preserving all accumulated knowledge so agents don't start from scratch each session. This is powered by two open-source tools:

  • entire-graph: Builds a precomputed repository map including ranked code searches, definitions, types, routes, and more.
  • entire-brain: Creates a persistent memory layer by combining the codebase map with past agent sessions and Git history.

The result is that agents always have "the right context at the right time, aka the 'why' behind the 'what changed.'"

Deeper Problem: Lost Reasoning#

A Hacker News commenter argued that engineers turned to AI because programming languages and tools are inadequate. Haimantika, working at Entire, reframed the issue: the real gap isn't tooling, but lost institutional knowledge. Commit messages fade, documentation rots, and nobody remembers why code exists. "Agents just made that acute, because now there's no one who remembers. That's the bit we're trying to fix."

The platform supports CI integrations with Buildkite and Depot, and runs on a fully distributed architecture that claims 89x faster pushes than GitHub. Entire is one of several new entrants reshaping source control, including Zed DeltaDB, Anthropic Scion, East River Source Control, Cursor Origin, and Radicle.

Sources