Meta has acquired Conductor, the startup whose software lets a developer run a team of AI coding agents at once, each working in an isolated copy of a codebase that the developer then reviews and merges. The deal was first reported by Axios on Wednesday, November 4, and Meta confirmed it to TechCrunch, although neither side disclosed the terms. Conductor was founded in 2024 and launched its core product in July 2025. It does not build its own models: the agents it orchestrates come from other labs, including Anthropic's Claude Code and OpenAI's Codex, and users bring their own subscriptions and keys. Conductor says more than 100,000 people now use the product.
The purchase is Meta buying the interface layer it has been missing. It spent much of the past year catching up in coding tools, launching its own Muse Code agent in August and recently reporting that its Muse agents have passed 10 million users. Conductor founder and chief executive Charlie Holtz, who started the company with chief technology officer Jackson de Campos, said the team has big plans for what he called the next interface and expects to have something new for customers by early next year. He will work with Alexandr Wang, Meta's chief AI officer, and Nat Friedman, who leads product at Meta Superintelligence Labs. Conductor raised a 22 million dollar Series A in March.
There is a price, and Conductor's existing customers are the ones paying it. The company has stopped new sign-ups and downloads, its cloud service becomes read-only on November 13 and shuts down entirely on November 20, after which it will begin deleting the data stored there. Cloud users must export their work or move it to their own machines before that date, and active subscribers will receive refunds. The local Mac application keeps working until early 2027, with new models and critical fixes ending as the team turns to its new project inside Meta. Conductor says it has never sold user data or trained models on it, and that it does not plan to start.
Meta was not the only lab buying coding infrastructure on Wednesday. OpenAI announced the same day that it will acquire Pierre Computer Company, the small team behind Code Storage and the open-source Trees, Diffs and Highlights libraries. Anyone who has browsed files or reviewed a change inside OpenAI's Codex application has already used Pierre's software, and OpenAI says Code Storage customers can keep using the service after the deal closes while its features are folded into Codex. Pierre's open-source projects will remain open source. That deal is subject to regulatory approval, the companies remain separate until it closes, and the price was not disclosed. Two days earlier, OpenAI's engineers had published a post about Pando, their in-house source-control backend built for humans and agents in the same repository.
Taken together, the two deals say something plain about where competition in AI coding has moved. The contest is no longer only about which model writes the best code. It is about who owns the surrounding machinery: where agents run, how their work is isolated and stored, and how a human reviews and merges what they produce. Conductor proved that a small team could build a valuable product entirely on top of rival models, which is exactly why a lab wanted to own it. Practitioners should draw the practical lesson. Orchestration tools built by startups are now acquisition targets, their shutdown dates can arrive with little warning, and any workflow that cannot be exported is a workflow that can be taken away.
