What if tens of thousands of AIs wrote code simultaneously? OpenAI's new challenge, 'Pando'

Imagine tens of thousands of programmers writing a single massive book at the same time. But these programmers are not human; they are AI agents writing hundreds of lines of code per second. They constantly save (commit) revisions and exchange data to ensure their work does not overlap. Could traditional software development infrastructure withstand this speed?

OpenAI recently introduced a new infrastructure called ‘Pando’ as the answer to this very question. OpenAI launches Pando: A Git backend for AI-scale development If traditional software development was human-centric, we now need a new road network suitable for the ‘agent era’ led by tens of thousands of AIs.

Why does this matter?

The core of software development is ‘collaboration.’ Every app and service we use is completed through a process where countless developers modify and merge code. Until now, a tool called ‘Git’ has been at the center of this process. However, the ‘agent-scale development’ that OpenAI pursues causes overloads that are difficult to handle with traditional methods. GitHub is rebuilding its Git infrastructure for AI agent-scale development

Agent-scale development means a method where numerous AI agents work in parallel with human developers, committing code at high speeds. OpenAI launches Pando: A Git backend for AI-scale development If the infrastructure cannot keep up with this speed, development stops, and the AI’s productivity becomes useless. OpenAI’s latest change suggests that AI is not merely writing code, but is directly paving the ‘highway’ for the entire software development ecosystem.

Understanding Pando: What is its role?

Git is, simply put, a ‘time machine and library for code.’ It records who changed what and when, and aggregates code written by multiple people.

To help you understand, let’s use an analogy. If traditional Git was a two-lane road, you can think of OpenAI’s Pando as a 16-lane superhighway where thousands of cars (AI agents) can travel simultaneously.

OpenAI developed Pando in just 10 weeks to provide a foundation for its coding agent, ‘Codex.’ OpenAI’s Pando Git backend reduced Git push latency for monorepos by approximately 80% Codex agents even participated in building the infrastructure during this process. Pando is designed to be compatible with existing Git while capable of handling massive code data reaching nearly one petabyte (about 1 million gigabytes) per day. OpenAI has created its own new Git backend

Current Status: How much faster is it?

The introduction of Pando has led to tangible improvements in development speed. According to OpenAI’s data, it has reduced intermediate push latency—the time it takes to save code to Git—by approximately 80%, or nearly fivefold compared to the traditional method. OpenAI has created its own new Git backend

This is not just a slight speedup. It means that situations where infrastructure used to paralyze when countless agents swarmed at once can now be handled comfortably. Based on this infrastructure, OpenAI is helping its coding agents solve more advanced and complex problems without human intervention. For instance, OpenAI recently deployed about 10,000 AI agents simultaneously to solve the ‘Navier-Stokes equations,’ a famous mathematical challenge, finding the answer in 88 hours. OpenAI claims to have solved a math problem that has stumped humans for decades

What happens next?

OpenAI’s recent move foreshadows a complete change in how software is created in the future. In the past, people typed on keyboards line by line to write code, but now an era is coming where AI agents tirelessly evolve software on top of massive collaborative tools (like infrastructure such as Pando).

We will frequently witness scenes where services and apps created by AI are updated much faster than before, and complex technical barriers are instantaneously broken through by AI. Of course, this process is accompanied by sophisticated monitoring systems to ensure that AI behavior does not deviate from the intended direction. OpenAI expands monitoring to detect unaligned behavior in frontier models

Pando has essentially strengthened the ‘muscles’ that allow AI to write code faster and more powerfully. The next time the app you use becomes noticeably smarter, keep in mind that tens of thousands of AI agents might be working night and day behind the scenes, traveling on the superhighway called Pando.

References

1. OpenAI launches Pando: A Git backend for AI-scale development
2. OpenAI has created its own new Git backend
3. OpenAI’s Pando Git backend reduced Git push latency for monorepos by approximately 80%
4. GitHub is rebuilding its Git infrastructure for AI agent-scale development
5. OpenAI claims to have solved a math problem that has stumped humans for decades
6. OpenAI expands monitoring to detect unaligned behavior in frontier models