Meta Buys Conductor: The Fight Isn't About the Model, It's About Who Orchestrates the Agents
Meta bought Conductor, the app that lets half a dozen coding agents work in parallel without stepping on each other. It didn't buy a model. It bought the desk where everyone else's agents live.
The deal surfaced on November 4: Axios broke it, founder Charlie Holtz confirmed it on the company blog, and Meta confirmed it to TechCrunch. The price wasn't disclosed, and it's worth saying up front because that's where speculation will grow.
What Meta Actually Bought
Conductor doesn't train models. Its job is to open several agents at once and give each one its own space.
An Agent Orchestrator, Not a Model
You launch Claude Code, Codex, Cursor and OpenCode from the same window, each in its own git worktree with its own diff panel. The orchestrator decides which agent does what, in which folder and on which branch, and keeps two sessions from touching the same file. The collision becomes a diff you review, not a corrupted file at three in the morning.
Read also
The Cloud Move: MicroVM Sandboxes That Spin Up in Seconds
The Cloud version runs in isolated microVMs with the repo and dependencies preinstalled. That lets an agent work for hours without occupying your machine or exposing your local environment.
Meta's Play for Coding Agents
Meta already had the bottom and the top of the stack. It was missing the middle.
Muse Spark 1.2 and Muse Code: the Model and the Terminal Agent
It has its own model aimed at coding agents, a terminal agent in beta, and the channel to sell them: the enterprise AI platform it launched in September.
What Was Missing Was the Window Developers Open Every Morning
The model is bought through an API. What's scarce is the habit: the app you open before typing the first line. Whoever owns that window owns which model gets used by default, even if switching is technically possible. And this isn't isolated: Meta brought in agent teams back in March, and other consolidations hit coding tools this year.
What Happens to Your Account and Workspaces
This is the only part that affects you today, and the announcement states it plainly.
The Local App Keeps Going Until Early 2027
The local app will keep receiving new models and critical fixes until early 2027. No shutdown date has been announced for Cloud, and it's better not to invent one: the post doesn't give it.
How to Move a Cloud Workspace to Your Mac
The changelog spells out the path: right-click the workspace and choose "Transfer to local". Repos and conversation history come with you, which is the difference between migrating and losing the work.
What the Announcement Doesn't Say: Price, Roadmap, Model Agnosticism
Three gaps: how much was paid, what happens to Conductor Cloud after 2027, and whether the app will keep launching competitors' agents once it belongs to Meta. Today it does. That it stays that way is confirmed by nobody.
Why This Matters Even If You Don't Use Conductor
Value Moved From the Model to the Orchestration
When any decent model is one API call away, the edge isn't the model, it's the layer that organizes work on top of it. That's what's being bought in 2026.
The Real Risk: A Workflow That Lives Inside a Closed App
If your decisions, your prompts and your project state live inside an app that can change owners, your continuity depends on someone else's roadmap. Before standardizing a team, three questions: can I export the full state, does it work with whichever model provider I pick tomorrow, and how long does someone need to replicate the workflow without the app?
What to Do This Week If You Depend on a Tool Like This
Everything that matters belongs in git: branches, commits, messages with context. If a chat history is the only explanation for why the code looks the way it does, that knowledge is at risk. The local-plus-cloud pattern Meta itself explored with local agents also works as a fallback.
And keep a plan B you can install in an afternoon: the CLI of your model provider plus your own review flow. It doesn't need to be better. It needs to exist, and you need to have tested it once before the day you require it.
Conclusion
Models are becoming interchangeable; work habits aren't. Whoever owns the window you launch agents from effectively owns the daily decision of which model you use.
For the average user the message is calm: the local app is supported through 2027 and there's a way to move workspaces. For whoever picks tools for a team, it's a reminder that portability gets designed up front, not negotiated afterwards. We keep covering these moves in the agent market here.
