Pando scales Git for coding agents

More coding agents mean more branches, pushes and repository downloads. OpenAI’s Pando disclosure shows how that activity can change the infrastructure underneath software development.

What happened

On 2 November, OpenAI described Pando, a backend for Git repositories that now supports development in its internal monorepo. It retains compatibility with the Git protocol and client.

The system is designed to handle many clients pushing to different branches while serving large amounts of repository data. OpenAI measured a nearly fivefold reduction in median push latency. Pando had already entered production; the announcement was an engineering disclosure rather than a new public product launch.

Why it matters

Writing code is only one part of a development workflow. Changes must also move through repositories, merge processes and integration systems without creating bottlenecks.

A backend designed for greater concurrency can help address those pressures while preserving familiar client tools. Its performance figures describe OpenAI’s deployment, rather than a guarantee for every repository.

The bigger picture

Agent adoption can shift engineering constraints into parts of the stack that were built around a different level of activity. Repository hosting is one example, alongside review and integration capacity.

Pando is useful as a concrete case study of that change. Broader availability remains separate from the internal deployment, so the disclosure should not be mistaken for a service that outside teams can already purchase.