OpenAI builds Pando, a Git-compatible backend for Codex-scale development

OpenAI says it has built a new backend called Pando that works with Git, the version control system most software teams use to track and share changes to code, while handling a volume of activity that ordinary Git setups struggle with. The company describes the project in a post on its site titled “Scaling Git at OpenAI.”

OpenAI ties the need for Pando directly to Codex, its AI coding agent. Codex can write, edit, and commit code far faster than human engineers working alone, which means more branches, more commits, and more history moving through a team’s version control system at once. Git was built for a pace of work set by people typing and reviewing code by hand. OpenAI says that once an agent is doing a large share of that work, the infrastructure underneath Git can become the limiting factor, not the engineers or the agent itself.

Pando is OpenAI’s answer to that bottleneck. By staying compatible with Git, it lets developers keep using the same commands, branches, and workflows they already know, while the backend handling storage and coordination underneath is built to absorb heavier and faster traffic. OpenAI frames the goal as removing infrastructure constraints so that large engineering teams can collaborate at the speed Codex enables, rather than being slowed down by tooling designed for a slower, human-paced era of software development.

What the announcement leaves out

The post does not include performance figures, a timeline for when Pando was built or put into use, or details on its internal architecture. It is also not clear from the announcement whether Pando runs only inside OpenAI’s own engineering organization or whether the company plans to make it available to outside developers or other companies. OpenAI does not say whether any part of the system will be released as open source, and the announcement does not name which teams at OpenAI use it today or how widely.

What OpenAI does make clear is the problem it set out to solve: as AI agents take on more of the actual writing of code, the tools built to manage that code need to keep pace too. Git has remained the standard for version control for close to two decades, largely because it was flexible enough to scale with human teams of almost any size. OpenAI’s framing of Pando suggests the company sees a different kind of scaling problem emerging now, one driven not by more engineers joining a project but by an agent capable of generating code faster than any individual engineer could. Whether Pando is a narrow fix for OpenAI’s own infrastructure or a preview of tooling other companies will eventually need is a question the announcement leaves open.