Conductor says it is joining Meta. The app, which lets developers run many AI coding agents side by side, announced the move in its own changelog on November 4, 2026. Axios, as relayed by Techmeme, reports that CEO Charlie Holtz says the team will join Meta Superintelligence Labs. The practical news for users is harsher than the headline: Conductor is winding down. Conductor Cloud goes read-only on November 13 and shuts down on November 20.
Meta has shown a pattern of buying agent tooling teams, from Stilla to the Manus deal that later resumed independent operations. This one is aimed at the developer workflow that Meta Muse Code is also trying to own.
TL;DR: what is confirmed
| Question | Answer | Source |
|---|---|---|
| Is the deal real? | Conductor says "Conductor is joining Meta" | Conductor changelog 0.93.0 |
| Where does the team go? | Meta Superintelligence Labs, per CEO Charlie Holtz | Axios via Techmeme |
| Price? | Not disclosed in sources we found | n/a |
| What happens to the product? | Wind-down "over the next weeks and months" | Conductor changelog |
| Cloud dates? | Read-only November 13, shut down November 20 | Conductor changelog |
| Local app? | New models and critical bug fixes until early 2027 | Conductor changelog |
| Data export? | Settings, Account, Export data; right-click workspace, Transfer to local | Conductor changelog |
What is Conductor?
Conductor is a coding-agent manager. Its site describes it as a way to "run a team of coding agents in the cloud," and says it runs "the first-party Claude Code, Codex, Cursor, and OpenCode agents under the hood." Users bring their own subscriptions and keys. The site claims "100k+ builders" and carries a user quote that it feels like "going from typing with two fingers to having eight arms." Those are Conductor's marketing numbers, not an audited figure.
The core idea is simple. A single agent works on one task and you wait. Conductor gives each agent its own isolated workspace, so you can start five tasks at once, review diffs as they finish, and merge the good ones. Locally, that is built on git worktrees. In the cloud product, each agent gets its own microVM sandbox. The site shows sandboxes booting as Firecracker microVMs with a repository snapshot restored, and says they "spin up in seconds so your agents can run for hours."
Recent features from the company's blog and changelog give a sense of the pace:
- March 30, 2026: a 22 million dollar Series A from Spark and Matrix.
- April 2, 2026: the team behind cmd joined Conductor.
- June 15, 2026: a note that Anthropic had delayed subscription billing changes for third-party Claude tools, so Claude plans kept working in Conductor.
- September 29, 2026: sign in with ChatGPT, so Codex can run on a ChatGPT plan. That tied into OpenAI's own login push, which we covered in Sign in with ChatGPT at DevDay.
- October 23, 2026: version 0.92.0 added remote control of your Mac's local workspaces from the Conductor iOS app.
- November 4, 2026: version 0.93.0, "Conductor is joining Meta."
What happens to users, and when
Conductor's changelog is specific, and it is the best primary source we have:
- Conductor will be wound down "over the next weeks and months."
- November 13, 2026: Conductor Cloud becomes read-only.
- November 20, 2026: Conductor Cloud shuts down.
- Early 2027: the local app stops getting new models and critical fixes.
If you use cloud workspaces, move them now. The changelog says to right-click a cloud workspace and choose Transfer to local, which moves the workspace and its chats to your Mac. For a full backup, use Settings, then Account, then Export data. Support is by email, listed on the changelog.
Local-only users have more time, but the direction is clear: the product has an end date. Do not build team processes around it.
Why would Meta buy this?
Meta has not explained its reasons in the sources we read, so this section is our reading, not Meta's statement.
Conductor's value is not its code alone. It is a team that has spent a year learning how developers actually run many agents at once: review loops, sandboxing, switching between models from different labs, and the interface for it. That is also the hard part of making a coding product that competes with Claude Code and Codex. The Conductor site itself shows a workflow that mixes models, for example asking a GPT model to review a proposed fix and starting a new session with another lab's model to rethink it. A team that has built a neutral manager over rival models now works for a lab that sells its own models.
It also fits a talent-and-team pattern. Axios's framing, per Techmeme, is that the team joins Meta Superintelligence Labs. Terms, headcount and whether Meta takes any code or only people are not stated in the material we have. Treat any price figure you see on social media as unverified until Meta or Conductor confirms it.
What is not known
- The deal price and structure.
- Whether Meta will ship a Conductor-like feature inside Muse Code, or a new product.
- Whether Conductor's support for rival labs' agents (Claude Code, Codex, Cursor, OpenCode) will survive in whatever Meta builds. Conductor says it runs those agents with the user's own subscriptions; a Meta product may not.
- Whether the local app's open or closed status changes. The changelog speaks only of fixes and models until early 2027.
We could not open the Axios article directly, which returned an access error. We cite it only through Techmeme's summary, and everything else here comes from Conductor's own pages.
Alternatives if you depend on Conductor
You do not need an app to run parallel agents. Options, from lowest to highest overhead:
| Approach | Good for | Trade-off |
|---|---|---|
| Plain git worktrees plus a terminal per agent | Anyone comfortable with git | No dashboard, manual review |
| Claude Code desktop parallel sessions | Claude Code users | Tied to one agent; see our parallel sessions and worktrees guide |
| Meta Muse Code multi-agent workflows | Teams already trying Meta's models | Newer; see Muse Code out of beta |
| Another agent manager app | Mixed-model workflows | Check each tool's data and billing terms |
A short worktree recipe works on any machine:
git worktree add ../feature-a -b feature-a
git worktree add ../feature-b -b feature-b
# start one agent per directory, then review diffs with git diff main
Whatever you choose, treat parallel agents as untrusted workers. Give each its own branch, review every diff, and keep secrets out of sandboxes. For agent guardrails, AgentBeam, from the explainx.ai team, is an agent security platform that stops AI agents before they take dangerous actions.
What people are asking
Is my code at risk? Conductor's changelog says nothing about sending data to Meta. It says the opposite of a handover in practice: it offers export and local transfer, and the local app runs on your Mac. If your repositories sit in Conductor Cloud snapshots, export your data and delete the account when you finish, and ask the support address in the changelog how snapshots are removed. We found no statement on retention, so ask directly.
Will my Claude or ChatGPT plan still work? Until the shutdown dates, Conductor says users bring their own subscriptions and keys. Its June 15 note said Claude plans kept working in Conductor after Anthropic delayed billing changes for third-party tools, and its September 29 post added Codex on a ChatGPT plan. Those arrangements belong to the old product. They say nothing about what Meta will build.
Is this an acqui-hire? The public wording is that the team is "joining" Meta, and the product is winding down. That looks like a team deal rather than a product purchase, but neither company has used the word acqui-hire in the material we read, and no terms are public.
Should I stop using parallel agents? No. The workflow is useful whichever app hosts it. The risk is lock-in to one manager, not the pattern itself.
What to do this week
- If you use Conductor Cloud, export data and transfer workspaces before November 13.
- Copy any custom rules, prompts and setup scripts into your repository.
- Pick a replacement from the table above and trial it on one real task before the cloud closes on November 20.
- Watch for a Meta announcement. If Meta ships a coding-agent manager inside Muse Code, we will cover it as its own story.
What this says about the agent-tooling market
Three signals stand out. First, labs prefer to buy the team that understands the workflow rather than build it alone: OpenAI took Pierre, a company working on Git for agents, as we covered in OpenAI to acquire Pierre Computer: reported vs. confirmed, and now Meta takes Conductor. Second, independent managers that sit on top of other labs' agents are fragile, because the labs can change billing terms, as the June note on Claude plans shows. Third, the shutdown dates are short. Nine days from the announcement, a cloud product goes read-only.
For developers, the lesson is portability. Keep your prompts, rules files and task specs in your repo, not only inside a tool's chat history. Then switching managers costs an afternoon, not a project.
Related reading
- Meta buys Stilla to link businesses to agent tools
- Meta Muse Code out of beta: multi-agent workflows
- Meta Muse Code launch
- Manus resumes independent operations
- OpenAI to acquire Pierre Computer
- Claude Code desktop parallel sessions and worktrees
Accurate as of November 4, 2026. Meta had not published its own statement in the sources we read; dates come from Conductor's changelog and may change.