Former GitHub CEO launched a distributed Git network for the era of agentic coding

In July 2026, the former CEO of GitHub launched an AI platform for code development — a distributed code management network designed from scratch for a world where AI agents are becoming the primary developers.

Why existing Git cannot handle agents

The current version control architecture — Git at its core, GitHub or GitLab as platforms — was designed for human work rhythms: one developer thinks, writes code, makes a commit, opens a pull request, waits for review.

AI agents work fundamentally differently. They are capable of simultaneously exploring dozens of approaches, instantly creating parallel branches, and generating commits at frequencies that overwhelm traditional workflows. When an agent writes thousands of lines of code in an hour, no team of humans can review each commit at the pace that the current development model assumes.

Centralized platforms scale poorly for parallel work of multiple agents on a single repository: merge conflicts, CI/CD pipeline bottlenecks, access level blocking — all of this was designed with the assumption of asynchronous human work, not synchronous machine work at industrial scale.

The startup proposes rethinking the infrastructure itself: not adapting existing tools to agents after the fact, but creating a system where agent workflows are native from day one.

Who is behind the new startup

The founder served as CEO of GitHub — a platform used by more than 100 million developers worldwide. Such a background provides a rare combination: a deep understanding of the limitations of existing infrastructure and enough authority to attract engineers and investors in the early stages.

Remarkably, the person who directly shaped the modern standard of collaborative development himself bets that this standard needs systematic replacement — not just another AI enhancement on top of the familiar git-flow.

Why is this happening right now

In 2025–2026, agent programming has moved out of the category of laboratory experiments. Systems such as Devin, Copilot Workspace, and dozens of analogues are already embedded in the workflows of technology companies. AI agents don't just offer autocomplete — they independently open branches, create pull requests, and conduct initial code reviews.

When there are more agents in a repository than people, current versioning tools become a bottleneck — and the infrastructure question ceases to be theoretical.

What this means

The exit of a GitHub veteran into the space of basic version control infrastructure is a signal: the market for agent coding tools is beginning to split into two layers. The first — agent systems themselves (Copilot, Cursor, custom LLM-agents). The second — infrastructure beneath them: repositories, synchronization networks, access protocols. Until now, the second layer remained in the shadow of the first; now people with strategic experience and understanding of what exactly needs to be built are entering it.

Why doesn't Git work for AI coding agents?

Git was designed for human work rhythm: developer thinks, writes code, makes a commit, waits for review. AI agents work fundamentally differently — able to work continuously and in parallel, requiring a new generation of infrastructure.