Meta Buys Conductor, and the Cloud Shuts Down in Two Weeks

Meta has acquired Conductor, the startup that lets engineers run a fleet of coding agents in parallel workspaces. Terms were not disclosed. The practical consequence landed faster than the strategy: Conductor stopped taking new sign-ups the same day, its cloud goes read-only on November 13, and it shuts down on November 20, after which all cloud workspaces and chats are deleted.

What Meta actually bought

Conductor launched in July 2025, when running a team of coding agents at once still sounded exotic. It raised a $22 million Series A in March from Spark and Matrix and became a familiar sight on engineering teams that wanted one agent per task, working in isolated branches, instead of one chat window doing everything.

The agents were never Conductor's own. The platform orchestrated models from Anthropic and OpenAI, which is exactly the layer Meta wants. Meta has its own coding agent, Muse Code, launched in August, and its Muse agents recently passed 10 million users. What it lacked was the workspace layer that makes several agents manageable on a real codebase.

Founder Charlie Holtz said the team has "big, crazy plans for the next interface" and promised something new by early next year. He is joining Meta's Superintelligence Labs under Alexandr Wang and Nat Friedman.

The shutdown clock is short

For anyone running Conductor today, the timeline is the story. Free users cannot upgrade, pro features are free for the remaining weeks, and active subscribers get refunds. Cloud workspaces transfer to a local Mac by right-clicking a workspace and choosing "Transfer to local"; full archives live under Settings, then Account, then Export data.

Local Conductor keeps getting new models and critical bug fixes until early 2027, then winds down and deletes all user data. The company says it never sold or trained on user data and will not start now.

Why orchestration is the prize

Parallel agent workspaces look like a convenience until you consider what breaks without them: conflicting edits, review queue chaos, and no clean boundary between what an agent attempted and what landed on main. Conductor solved a workflow problem, not a model problem, and workflows are not something a lab can fix by training a bigger model.

That makes this a pattern. Meta bought the orchestration surface, not the intelligence on top of it. The same logic explains why tools that coordinate agents keep getting acquired while single-model wrappers do not: the differentiator is the harness that decides which agent touches which file.

What it means for builders

Three things. If Conductor Cloud sits in any pipeline, export now, not on November 19. Treat any hosted agent-orchestration service as a migration risk and keep a local or self-hosted fallback for the workspace layer. And expect parallel agent workspaces to reappear inside Muse, which is what an acqui-hire of this shape is for.

(Related: Uber's first agent numbers show how thin real agent adoption still is outside code.)