OpenAI logo (Credits: OpenAI)
OpenAI has rebuilt the engine room of its engineering org. The company just revealed Pando, a brand-new backend for Git repositories designed to survive the agentic-scale development that its Codex coding agents unleashed — cutting median push latency by nearly 5x and now delivering nearly a petabyte of Git data per day.
Why OpenAI Built Pando to Scale Git
Over the past year, the size of OpenAI’s monorepo and the demands on it grew “at a staggering pace,” according to Aaron Friel, Taylor Blau and Ted Nyman, the members of technical staff behind the project. Codex changed the scale and pace of development — one engineer now has several changes ready for CI at once, each push can trigger multiple jobs fetching the same commit, and the volume of branches, pushes, merges, fetches and clones exploded.
By the end of last year, the old system was struggling: slow fetches and pushes, failed merges, and reliability at times falling below 90%. Rather than abandon Git — which grew out of a community exchanging kernel patches by email two decades ago — OpenAI chose to keep the Git protocol and rebuild the backend underneath it.
Pando: Horizontally Scalable Writes, Git-Compatible
Pando was designed around two goals: horizontally scalable writes optimized for concurrent pushes to many branches, and maximum read throughput so clients can download data fast enough to saturate their network bandwidth.
Crucially, Pando retains full compatibility with the Git protocol and client. That meant no retraining, no workflow changes, and seamless adoption for both OpenAI’s engineers and its AI agents. The result: Pando now powers all development in OpenAI’s monorepo without interrupting any engineering work during the transition.
The move is a signal for the wider industry: as coding agents multiply the pace of software development everywhere, the tooling underneath — source control included — has to scale with it. OpenAI says it will continue contributing to the Git ecosystem to support agent-scale work.
Also Read: AgentGrid.io Launches as a Shared Workspace for Humans and AI Agents to Build on the Same Live Files | Meet Anthropic’s Sandbox-Runtime: OS-Level Sandboxing for AI Agents, No Containers Needed
Source: OpenAI — read the full story there.