On Wednesday, November 4, the two biggest names in AI made the same move within hours of each other. OpenAI announced it will acquire Pierre Computer Company, a small startup whose main product is Code Storage, a service for storing code. Meta, meanwhile, confirmed to TechCrunch that it has acquired Conductor, an app that lets developers run several AI coding agents at the same time. Neither company is best known for filing cabinets or traffic control. That is exactly the point.
A day later, Cat Wu, the product leader behind Claude Code at Anthropic, announced she is joining OpenAI. A storage deal, an agent manager deal, and a headline hire from the rival camp all landed in the same week. Together they show where the fight over AI coding agents is moving: away from the act of writing code, and toward everything that happens after it is written.
Git, branches and agents, in plain English
Almost all professional software is kept in a system called Git. Think of Git as a shared document folder with a perfect memory. Every change to the code is recorded, nothing is ever lost, and any earlier version can be pulled back.
Git also lets people work in parallel through branches. A branch is a private copy of the project where one person, or one program, can experiment without breaking the main version. When the work looks good, another developer reviews it and merges it back into the main line. Storage, review and merging are the quiet plumbing of software, and they mostly run through services built on Git, above all GitHub, which Microsoft owns.
Coding agents are AI programs that do not just suggest lines of code but carry out whole tasks: fixing a bug, building a feature, running tests. Tools such as OpenAI’s Codex and Anthropic’s Claude Code can already write large amounts of working code. The bottleneck is no longer typing. It is keeping track of dozens of agent-made branches, checking what each agent changed, and deciding what is safe to merge. That is the layer OpenAI and Meta just bought into.
OpenAI buys the storage layer
Parts of Pierre are already inside OpenAI’s product. In its announcement, OpenAI said Pierre’s open source libraries Trees and Diffs already power the file browsing and change review that Codex users see when an agent finishes a task. Its commercial product, Code Storage, goes further: it is a Git based storage service with controls designed for programs, not people, so apps can store versions and fetch history on their own.
“We want to make Codex the place where developers and agents take on their most ambitious projects,” OpenAI said in the announcement. The price was not disclosed, and the deal still needs regulatory approval, with the two companies staying independent until it closes. OpenAI said Code Storage customers can keep using the service after the deal closes while its features are folded into Codex, and Pierre’s open source projects will stay open source.
Two days before the announcement, OpenAI had published an engineering post about Pando, its own in-house Git backend. Read together, the message is hard to miss: OpenAI is rebuilding the storage layer for a world where most new code is written by machines, and it would rather own that layer than rent it.
Meta buys the orchestration layer
Conductor attacks the problem from the other side. Instead of storing code, it manages the workers. The app lets a developer split a project into separate workspaces and set a different coding agent loose in each one, running in the cloud, locally or from a phone. Conductor does not make its own AI models: the agents it coordinates come from labs such as Anthropic and OpenAI, according to TechCrunch.
“I have big news: Conductor is joining Meta,” founder and chief executive Charlie Holtz wrote on the company blog on Wednesday. Meta confirmed the acquisition to TechCrunch but shared no price or further details. Conductor was founded in 2024, so Meta is buying a very young company for its team and its approach. Existing users have a deadline to note: Conductor said its core features will keep working until early next year, but its cloud service shuts down on November 20, when it will begin deleting user cloud data.
Meta is not coming to coding cold. It launched its own coding agent, Muse Code, in August. What it lacked was a control room for running many agents at once. Conductor is that control room.
Why the boring layer is the prize
There is a pattern in technology: when a tool becomes popular, the money and power sit in the layer everyone has to pass through. For software, that layer is where code lives and gets approved. Whoever runs it sees every project, sets the rules for how agents connect, and becomes very hard to replace.
The week’s third move underlines the point. Cat Wu ran product for Claude Code, Anthropic’s coding tool and Codex’s closest rival, and she announced on X that she is joining OpenAI to lead product management for its Core Products group, working with Thibault Sottiaux, the executive who runs Codex. OpenAI confirmed the appointment. Wu joined Anthropic in August 2024 and, with Claude Code creator Boris Cherny, was one of the product’s two public faces, Business Insider has reported. Talent, storage and orchestration are all being pulled in house at once.
What should developers watch next? Three things. First, whether Code Storage features start appearing inside Codex, turning it from a coding assistant into the place where code lives. Second, what Meta builds with the Conductor team, and whether its agent tools stay open to rival models or close around Meta’s own. Third, how GitHub responds, because every one of these moves points at ground it has held for years. The agents learned to write the code. Now the race is to own the desk where the code is filed, checked and signed off.