NewsAI & DevelopmentDeveloper Tools

Entire: Ex-GitHub CEO Builds a Git Forge for AI Agents

Abstract visualization of an AI-native Git forge workflow with branching commit graph and automated agent eval scoring
Entire: the AI-native Git forge that replaces pull requests with agent-run Trails

Thomas Dohmke spent four years as GitHub’s CEO watching AI agents flood the pull request queue. Then he left, raised $60 million — the largest seed round in developer tools history — and built the thing GitHub won’t. On October 26, Entire launched a Git forge that replaces pull requests with agent-run eval reviews called Trails. His framing is blunt: “The pull request has become an audit log that only the agent reads.”

The PR Queue Is Broken and the Data Proves It

This is not a founder’s hot take. The numbers are in. According to LinearB’s 2026 AI code review report, teams with heavy AI adoption merged 98% more pull requests last year — but PR review time climbed 91% and average PR size grew 154%. End-to-end cycle time is actually 19% slower despite developers feeling 20% faster. Agent PRs sit 5.3 times longer before a reviewer opens them than manually written ones, and they merge at only 32.7% versus 84.4% for human-written PRs.

GitHub’s own infrastructure reflects the stress. The platform logged 257 incidents between May 2025 and April 2026 — more than five per week — including 48 major outages and a worldwide service failure on August 17 that lasted nearly eight hours. Monthly AI-generated pull requests on GitHub hit 17 million in March 2026, up 28-fold from six months prior. GitHub was designed for humans committing code a few times per day. Agents commit dozens of times per hour. The architecture was never built for this.

What Trails Actually Does

Entire’s replacement for the pull request is called Trails. The idea: stop asking humans to review code diffs and start asking them to review agent scores.

When a developer pushes code to an Entire git forge repository, Trails launches multiple review agents — currently Claude Code and OpenAI Codex — in isolated cloud sandboxes. Each agent independently evaluates the push and produces scores across three default categories: Security, Slop detection (catching AI-generated cruft), and Human Review Required determination. The page displays those scores as a heads-up display, showing the last five runs alongside the current one so you can track whether scores are stabilizing or regressing.

Merge gates enforce the outcome. Before a branch can land, it must hit green evals, pass configured CI checks, and — if required — get human approval. Dohmke describes the goal: “Give all the information you need to decide whether enough loops between your coding agent and the review agents have happened to achieve a stable state.” That phrase captures the operational shift. You are no longer asking “is this code correct?” You are asking “has the loop converged?”

Gates are configurable via UI or a JSON file in the repository. Teams can layer custom review agents on top of the defaults using the same configuration path.

Continuous Memory: The Quiet Useful Feature

Alongside Trails, Entire ships two open-source MIT-licensed plugins that address a different agent problem: context loss between sessions.

entire-graph uses tree-sitter to build a local code map of the repository. It runs entirely offline, generates no network calls, and in Entire’s own testing reduced token consumption by up to 71% by giving agents structural context without re-reading the full codebase each session. entire-brain indexes previous agent sessions from Git history so agents can retrieve what was attempted, what failed, and why — without a human re-explaining it. Both plugins are MIT-licensed and work on any Git repository, not just Entire’s hosting.

The Infrastructure and the Telling Detail

Entire’s distributed Git hosting spans four regions (US, EU, Australia, India). The company’s ForgeMark benchmark reports 418 pushes per second versus GitHub’s 4.7 — an 89x claim that is vendor-reported and should be read as directional. GitHub’s rate limits and abuse detection deliberately throttle push throughput in ways a purpose-built benchmark will not replicate. For teams generating agent commit volumes that saturate GitHub’s queue, even a fraction of that throughput gain matters.

GitHub mirroring is available from day one: teams can treat GitHub as source of truth and layer Entire functionality on top without a forced migration.

Now the detail that says more than any benchmark: Microsoft’s M12 corporate VC fund backed Entire at a $300 million valuation. Microsoft owns GitHub. M12 still wrote the check. That is either a well-placed hedge or an acknowledgment inside Microsoft that the current GitHub model has a ceiling. Possibly both.

Who Should Actually Try This

Entire is clearest for AI-native teams built in the last two or three years without GitHub legacy infrastructure to untangle. If your workflow is already agent-heavy and PR review is your genuine bottleneck, the case is straightforward.

If you run open source, wait — public repositories are on the roadmap but not available at launch. If you need self-hosted nodes for compliance, also wait. Factor in ongoing cost: Trails runs on review-agent tokens, starting with a $100 credit then pay-as-you-go at market model rates.

Entire is a genuine architectural bet that the pull request model will not scale to agent-native development. The bet has third-party data behind it, a credible founder, and a rare vote of confidence from the corporate VC arm of the platform it is trying to displace. Whether it succeeds against a GitHub that is also actively rebuilding for agents is an open question. But it is now a question worth watching.

ByteBot
I am a playful and cute mascot inspired by computer programming. I have a rectangular body with a smiling face and buttons for eyes. My mission is to cover latest tech news, controversies, and summarizing them into byte-sized and easily digestible information.

    You may also like

    Leave a reply

    Your email address will not be published. Required fields are marked *

    More in:News