AI Coding Startup From Former GitHub CEO Rebuilds Version Control for Agents

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

Why existing Git struggles with agents

Today's version control architecture, with Git at its core and GitHub or GitLab as the platform, is built around a human rhythm of work: a developer thinks, writes code, makes a commit, opens a pull request and waits for review.

AI agents work in a fundamentally different way. They can explore dozens of approaches at once, create parallel branches instantly and generate commits at a frequency that overloads traditional workflows. When an agent writes thousands of lines of code in an hour, no team of humans can review every commit at the pace the current development model assumes.

Centralized platforms scale poorly when many agents work on a single repository in parallel: merge conflicts, bottlenecks in CI/CD pipelines and access-level locks were all designed for the asynchronous work of people, not the synchronous work of machines at industrial scale.

The startup proposes rethinking the infrastructure itself: rather than retrofitting existing tools for agents, it wants to build a system where agentic workflows are native from day one.

Who is behind the new startup

The company's founder served as CEO of GitHub, a platform used by more than 100 million developers worldwide. That background offers a rare combination: a deep understanding of the limits of existing infrastructure and enough authority to attract engineers and investors at an early stage.

Notably, the person who directly shaped the modern standard for collaborative development is betting that this standard needs a systemic replacement, not yet another AI layer on top of the familiar git-flow.

Why this is happening now

In 2025–2026, agentic programming moved out of the lab. Systems such as Devin, Copilot Workspace and dozens of similar tools are already embedded in the workflows of tech companies. AI agents do not merely suggest autocompletions: they open branches on their own, create PRs and perform first-pass code review.

When agents in a repository outnumber humans, current versioning tools become a bottleneck, and the question of infrastructure stops being theoretical.

What this means

A GitHub veteran moving into foundational version control infrastructure signals that the market for agentic coding tools is starting to split into two layers. The first is the agentic systems themselves (Copilot, Cursor, custom LLM agents). The second is the infrastructure beneath them: storage, synchronization networks, access protocols. Until now the second layer has stayed in the shadow of the first; now people with strategic experience, and an understanding of what exactly needs to be built, are moving into it.

Why is Git not suited to AI coding agents?

Git is designed around a human rhythm of work: a developer thinks, writes code, makes a commit and waits for review. AI agents work in a fundamentally different way. They can run continuously and in parallel, which calls for next-generation infrastructure.