Room for more builders.
A coding assistant can help write a change quickly. A team still needs somewhere reliable to save that change, share it and combine it with everyone else’s work. OpenAI’s latest engineering report tackles that less glamorous—but essential—part of the job.
In a November 2 account of its Pando system, OpenAI says the new software now powers all development in its main shared code repository. The company reports that switching to Pando reduced the wait for a typical code upload, measured by median push latency, by nearly 80% and improved merge reliability.
Those are results from OpenAI’s own environment, not a promise of the same speedup for every software team. They offer a concrete example of infrastructure adapting to engineers and coding agents working on more changes at once.
Keep the familiar tools.
Git is the version-control system developers use to record code changes and coordinate work. A “push” sends local changes to a shared repository; a “fetch” brings shared history back. Pando replaces the machinery serving that repository while remaining compatible with Git’s protocol and client.
OpenAI describes two practical design choices. Changes to different branches—separate lines of development—can be handled independently. And the service reuses compressed bundles of repository history instead of preparing a fresh bundle for each reader. That helps many developers and agents retrieve the same material with less repeated server work.
The design has a specific tradeoff: readers checking several branches do not necessarily see one synchronized snapshot across the entire repository. Updates to each individual branch remain coordinated. The familiar workflow stays; the machinery behind it changes.
Built with coding agents.
The company says it built Pando with Codex in ten weeks: one engineer designed the core, teams of agents implemented it, and a handful of engineers deployed it. Production traffic began in April, so the new report describes an operating system rather than a just-announced development plan.
OpenAI also describes checks against standard Git behavior and simulations to find subtle coordination problems. People designed, tested and deployed the service alongside the agents. The useful story is a working collaboration, not a claim that software engineering no longer needs engineers.
Useful work beyond one company.
OpenAI says some improvements made along the way are already in upstream Git, with others in its public Git repository. It is exploring ways to share Pando technology with users; the report does not announce a generally available Pando service or release the full backend.
Our reading: SI-assisted development becomes more useful when the surrounding tools keep pace. Faster sharing and dependable collaboration give teams more room to turn ideas into working software. Pando is a tangible example of that supporting work—and contributions to Git offer a route for benefits to reach beyond one lab.
