OpenAI Agrees to Buy Pierre Computer, a Roughly 10-Person Git Startup, Two Days After Detailing Pando
A.I. / news
OpenAI Agrees to Buy Pierre Computer, a Roughly 10-Person Git Startup, Two Days After Detailing Pando
The price is undisclosed. The deal and OpenAI's November 2 post on its Pando backend point the same way: Git is straining under parallel coding agents.

OpenAI agreed on November 4 to acquire Pierre Computer Company, a San Francisco developer-tools startup whose Git storage layer, Code.Storage, it plans to fold into Codex. Neither company disclosed a price, and the deal is subject to customary closing conditions including regulatory approvals.
InfoWorld said Pierre's headcount is "just barely in the double digits," or roughly ten people. The law firm Goodwin, which advised Pierre, confirmed the agreement in a November 4 announcement naming Joshua Rottner, Matthew Baudler and Steve Wheeler as its deal team leads.

What OpenAI is buying
Pierre makes Code.Storage and two open-source libraries, Trees and Diffs. OpenAI already uses the libraries in the Codex app to browse files, edit code and review changes, according to Goodwin. InfoWorld said OpenAI pointed to Diffs Edit mode and a feature called Highlights.
Goodwin quoted Pierre's own description of its work as "low-level, composable product infrastructure for machine work."
InfoWorld said OpenAI will keep Pierre's open-source projects open and keep contributing to them. Code.Storage customers can keep using the product until its features are brought into Codex. No timeline for that move has been published.
Pando: the November 2 post
Two days before the deal, OpenAI published "Scaling Git at OpenAI," describing Pando, an internal Git backend. Developers Digest, summarising the post, said Codex-driven parallel work brought more branches, pushes and merges. It said that from around the end of 2025 the monorepo systems had slow fetches and pushes and failed merges, with reliability at times below 90%.
The figures below are OpenAI's own, as relayed by Developers Digest. No independent party has verified them.
| Pando measure | Figure |
|---|---|
| Median push latency | Cut by nearly 5x, or nearly 80% |
| Git data served per day | Nearly 1 petabyte |
| Packs served in one hour | More than 57 TB |
| Cache hit rate | Above 99% |
One engineer designed the core and teams of Codex agents implemented it, over 10 weeks, per the post. Production traffic began in April, and Pando now powers all development in OpenAI's monorepo.
Pando orders updates per branch instead of across the whole repository, so pushes to unrelated branches do not conflict. Only pushes to the same ref do. It stays compatible with the Git protocol and clients. The post's wording on the trade-off: "users and agents are agnostic to the global ordering of ref updates."
What is not yet known
Pando is internal and cannot be installed. Developers Digest said OpenAI is exploring ways to share the technology, which is not a commitment, and no customer launch has been announced.
The sources do not say whether Pierre's team will work on Pando, whether Code.Storage will become a paid Codex feature, or what OpenAI paid. InfoWorld quoted OpenAI's announcement: "With the Pierre team, we wish to set the standard for modern software development."
Coding agents are changing what developer tools have to handle. A Copilot update for Xcode 27.3 and JetBrains Air and one developer's account of leaving Claude Code over 244 releases are earlier examples of that churn on The Terminal.
OpenAI has not published a closing date for the Pierre acquisition.
Sources
More in A.I.
- 01Google's Gemini Help Page Says Human Reviewers See Drive and Gmail ExcerptsA GitHub post from a user named thisisme6598 claims Gemini reads files unprompted. Google's own page confirms review, but not that part.
- 02Anthropic Ships Claude Fable 5.2 at $8 per Million Input TokensArtificial Analysis ranks the model first of 74 on its index, but its run needed 275 million output tokens and 44 minutes per task.
- 03OpenAI's GPT-6 Cyber Scores 48.45% on ExploitGym, but Only Vetted Daybreak Red Customers Get ItThe system card shows a smaller gain over GPT-6 Astra than the headline numbers suggest, and lists what the model still cannot do against hardened targets.
- 04HyperFlow H3 Flash Shrinks MiniMax-H3 From 31B to 13.7B, but Its Licence Excludes the US, EU and UKVideo Rebirth's compressed video-and-audio model runs in 6 steps and fits a 24 GB GPU in GGUF. The speed figures are the developer's own, and no quality benchmark is published.