Meta has acquired Conductor, a startup founded in 2024 whose Mac app lets engineers run several coding agents on the same project at once. Axios reported the deal first and Meta confirmed it to TechCrunch; the price was not disclosed. Conductor founder and CEO Charlie Holtz wrote on Wednesday that the team was joining Meta and that it had "big, crazy plans for the next interface."
What Conductor built was unglamorous and useful: give one agent a task, give a second a different one — a feature here, tests there — let each work in its own copy of the repository, and have the human review and merge. The important part was neutrality. The agents it orchestrated were not its own; they came from Anthropic (Claude Code), OpenAI (Codex) and Cursor. Conductor was the control room above its competitors' tools.
That control room is now being shut. New signups and downloads have stopped. Conductor Cloud goes read-only on November 13 and offline on November 20, after which the company begins deleting the user data stored there; customers can export or move their workspaces onto a Mac until then. The local app will keep receiving new models and critical bug fixes into early 2027.
For Meta this fills a gap in a fairly short list of homegrown coding products. The company launched Muse Code, its own coding agent, in August, and recently reported more than 10 million users across its Muse agents. Holtz praised working with Alexandr Wang, Meta's chief AI officer, and Nat Friedman, head of product at its Superintelligence Labs — "the highest talent density of all the labs," he wrote on X.
The timing is not incidental. The same day, OpenAI told Axios it plans to acquire Pierre Computer, a startup that stores and manages AI-generated code. In June, OpenAI had already announced Ona, which specialises in cloud coding environments and agent orchestration. Within a year, the big labs have bought the plumbing underneath the agent: the queue, the repository, the runtime.
That is a shift in what is being competed over. While models differed mainly on benchmarks, the tooling layer was a neutral market. Once a lab owns that layer, it has a say in which model is the default and which is the alternative — a far more effective form of lock-in than a good benchmark score.
The open question follows directly. Conductor's reason to exist was routing work to Claude Code and Codex. Under Meta it can hardly keep doing that unchanged without subsidising rivals. Its promise of a "next interface" for early next year points more towards a Meta product that defaults to Muse than towards a neutral marketplace, and nobody at Meta has said otherwise yet.
For users there is a 15-day deadline and a lesson the industry is currently learning on a monthly cadence: if you build your workflow on a startup's cloud, you are building on its balance sheet. Conductor at least offered exports and a local app, which is more than most acquihire shutdowns manage. In March it had announced a $22 million Series A led by Spark Capital and Matrix, with participation reported from engineers at Meta, Google and Spotify.
What to watch is whether Meta ships an orchestrator that treats outside agents as first-class, or one that mostly scales its own model. For developers that is the real headline: what is being consolidated is not the model but the queue in front of it.
