Meta Acquires Multi-Agent Coding Tool Conductor: Cloud Shuts Down November 20, Data Deletion Begins
Meta acquired Conductor on November 4, 2026: the tool that runs multiple AI coding agents in parallel workspaces, without building models of its own. New sign-ups stopped the same day; Conductor Cloud goes read-only November 13 and shuts down November 20, when cloud data deletion begins. The local Mac app keeps getting updates until early 2027. The real deadline is the 13th, not the 20th.

The short version: the acquired tool's cloud service has days left
On November 4, 2026, Meta acquired Conductor, a multi-agent coding tool. Axios reported the deal first; Conductor founder and CEO Charlie Holtz confirmed it the same day on the company blog — "Conductor is joining Meta" — and Meta confirmed the acquisition to TechCrunch, without disclosing any deal terms, including the price.
For people still using Conductor, the acquisition itself is not the point. The timeline after it is: no new sign-ups or downloads from November 4; Conductor Cloud goes read-only on November 13; it shuts down on November 20, when deletion of cloud user data begins. As this article is written, the real deadline — November 13 — is one day away. Note that: it is the 13th, not the 20th. By the 20th the cloud service is already gone; the window to export data and transfer workspaces closes on the 13th.
The local Mac app gets more time: it keeps receiving new models and critical bug fixes until early 2027, then winds down gradually, with all user data deleted at that point. In other words, Conductor is counting down week by week — the cloud first, then local, then the data itself.
What Conductor is: no models of its own, just a way to run a team of agents together
Conductor does one thing, put simply: it lets developers deploy multiple AI coding agents into parallel workspaces to work on a project together. Each agent sits in its own isolated workspace, changing its own part, without stepping on the others, and the changes are brought together at the end.
It builds no models itself. The agents running on it come from the big labs — Anthropic, OpenAI, and others — Claude Code, Codex, Cursor, OpenCode can all be plugged in. What Conductor sells is orchestration and interface: how workspaces are divided, how progress is tracked, how changes are merged. It can run in the cloud, on a local Mac, or be deployed and controlled from a phone.

That positioning looks very representative in November 2026. Models themselves are less and less the scarce resource; the scarce part is how to get a group of agents to work together without colliding. Conductor bet on that layer, and Meta paid for exactly that layer — not a model, but a way of managing many of them at once.
Your phone works as a command deck too: remote control arrived in 0.92.0
Conductor runs in three places: the cloud, a local Mac, and your phone. The phone is not just for checking progress: version 0.92.0, released on October 23, 2026, added remote control, so an iPhone can browse workspaces, send messages to agents, and stop a running agent at any time.
That fills a gap multi-agent orchestration often ignores: nobody sits at a desk all day. Open a few parallel workspaces on a Mac during the day, then keep watching progress and answering messages from your phone in the evening; the agents keep working in the cloud whether or not you are at your computer. Plenty of parallel tools only solve "run more agents on one machine." Conductor also built in "stay in charge while you are away from it."
That is part of the orchestration ability Meta is buying: not just split screens and merged changes, but a team of agents that can be managed even when nobody is watching. The catch is that these phone-first users are now also the hardest to migrate — replacements do not necessarily offer the same phone experience, a point we come back to below.
The user timeline: four dates, and none of them can be mixed up
Conductor's official announcement lists the timeline clearly. It is reproduced here in full, because each date means a different thing:
- November 4 (already happened): no new sign-ups or downloads. Existing free users can no longer upgrade to Pro, but Pro users keep Pro features for free.
- November 13 (Friday — the real deadline): Conductor Cloud goes read-only. You can read old chats and export data, but you cannot send new messages. Active subscribers get a refund.
- November 20 (Friday): Conductor Cloud shuts down, and deletion of cloud user data begins. The official wording is "please export your data before this date."
- Early 2027: the local Mac app keeps getting new models and critical bug fixes until early 2027, then winds down gradually; all user data is deleted at that point. The company states it has never sold user data or trained on it, and the data will be deleted cleanly on shutdown.
The easiest mistake here is treating November 20 as the deadline. The 20th is the day the service shuts down and data deletion starts — not the last day to act. For cloud users, the window to actually do things ends the moment read-only mode begins on the 13th. One day late, and a workspace can only be viewed, not used.
Three things to do right now
If you are a Conductor Cloud user, do not wait. Do these three things, in this order:
First, move your cloud workspaces to your local Mac. The official method is ready-made: right-click a workspace and choose "Transfer to local," and the workspace moves to your Mac together with its chats. This is Conductor's own recommended path, and it preserves more than a file export does.
Second, export a full data archive. The path is Settings → Account → Export data. Even if you have already transferred your workspaces locally, export an archive as well — belt and braces. Exporting is still possible while read-only after November 13, but do not bet on the last minute.
Third, confirm your refund. Active subscribers will get a refund without having to apply, but it is worth checking your billing to confirm it arrived. Free users are unaffected, except that Pro upgrades have been unavailable since November 4.
If you only use the local Mac app, there is no short-term panic: new models and critical fixes continue until early 2027. But start thinking about an exit plan — this product has no long-term future, and after early 2027 even the local version winds down. Migrating gradually while it still works beats being forced to move later.
A company acquired in two years: founded in 2024, product live for barely over a year
Conductor's timeline is strikingly short. The company was founded in 2024; its core product only launched in the summer of 2025 — barely over a year before the acquisition. In March of this year it raised a $22 million Series A from Spark and Matrix, announced on its own blog on March 30. Less than eight months after the Series A, the company was folded into Meta.
That rhythm is not unusual in the AI coding-tool race. A small team proves out a way of working; a large company buys the team and the product whole, because it is faster than building it from scratch. Conductor's team is small — around ten people, according to secondary reporting — and founder Charlie Holtz joins Meta Superintelligence Labs, working with chief AI officer Alexandr Wang and product head Nat Friedman. Holtz wrote in the blog post that there are big plans for the next interface, with "something new" for users by early next year.
For founders, the story cuts both ways. On one side: a team barely two years old got an acquisition offer from a giant on the strength of one idea — orchestrating multiple agents — which would have been unthinkable a few years ago. On the other: a product live for barely a year is being shut down, and the cost of that trust and data migration lands entirely on users. The $22 million Series A ended up as a ticket into a large company, not fuel for the product to keep growing on its own.
Why now: one more item on Meta's acquisition list
This acquisition is not a one-off. TechCrunch's report notes that Meta has made a string of AI-related acquisitions over the past year, including Moltbook, the viral agent platform. Meta lagged behind in AI services and has spent the last several months catching up: it launched its own coding agent, Muse Code, in August; its Muse agents reportedly passed 10 million users; and now it has bought Conductor's entire team.
The pattern says Meta's approach has changed: it is no longer betting only on building one big model in-house, but on owning the whole chain — model, orchestration, and interface. Models can be chased with hiring and compute; the feel of orchestration and interface has to be earned from a real product. Conductor's team is only about ten people and its product is barely a year old, but what it built is exactly the piece Meta lacks most: getting multiple agents to divide work on one project without colliding.
No terms were disclosed, and Meta declined to say anything beyond confirming the deal. That silence is itself a signal: Meta is not buying the Conductor brand — the brand is about to disappear. It is buying the team and the "next interface" in their heads. Holtz's "big, crazy plans for the next interface" translates to: Conductor was only a prototype; the real thing is being built inside Meta.
Who gets hurt most: heavy cloud users, and anyone who made it their main workflow
The people hit hardest are those whose main workflow lives in Conductor Cloud. Local Mac users have at least a few months of buffer; cloud users are counting in days. Workspaces, conversation history, agent context — all of it is in the cloud, and once read-only mode starts on November 13, all that is left is exporting and transferring.
Next are the people who deploy and control agents from their phones. Conductor only added phone remote control in version 0.92.0 in October — an iPhone can browse workspaces, send messages, and stop agents. Users who turned Conductor into a pocket agent console face the highest migration cost, because a replacement may not offer the same mobile experience.
Least affected are light users who treated it as "several Claude Codes running side by side." That use case has the most replacements: open several terminal windows, split with tmux, or switch to another parallel-workspace tool, and a workable setup takes half a day. What is genuinely hard to move is a workflow where project management, conversation history, and review flow have all grown into Conductor itself.
How to pick a replacement: three things, before the feature list
With Conductor shutting down, the first instinct is to find something that looks like it. But three things matter more than resemblance:
First, who owns the data, and where does it live? The painful lesson of this shutdown is that a cloud workspace can vanish on short notice. When evaluating a replacement, ask first: can workspaces and conversation history be fully exported? Can data live on your own machine or in your own repository? A tool with no one-click full export will leave you panicking again at the next shutdown.
Second, is it tied to one model vendor? Conductor's biggest strength was neutrality: Claude Code, Codex, Cursor, OpenCode could all plug in. Prefer an orchestration layer that stays equally neutral over a vendor's own workbench that only runs its own agents. Models come and go; the more neutral the orchestration layer, the less you have to move when you switch models.
Third, team size and business model. Conductor raised a $22 million Series A and had a product live for over a year — and still shut down on short notice. When choosing a replacement, look at what keeps it alive: subscription revenue, or waiting to be acquired? No tool can promise to never shut down, but at least pick one with a dignified exit path — local-first, exportable data, open source, or a clear migration plan.
There is no need to rush the decision. Get your data out of Conductor safely first, then try replacements slowly. The migration window is short, but the choice of replacement can wait — the local app keeps working until early 2027.
If you never used Conductor, should you care?
If you never used Conductor, this still matters to you — just not in the "export your data now" sense. It is a weather vane: when even a niche need like "orchestrating multiple agents" is worth Meta buying a whole team for, multi-agent parallel development has moved from geek hobby to contested territory.
Over the next year, you will see the problems Conductor solved showing up in more tools: how multiple agents divide work, how changes get merged, who wins when they conflict. Some of it will grow inside Copilot, Cursor, and other large tools; some will be new standalone products. Conductor's shutdown is not the end of this direction — quite the opposite, it is the starting line after a large company stamped its approval and cleared the field. When Meta shows its "something new" early next year, compare it against what Conductor looked like today: which interactions survived, and which were thrown away.
One detail is worth recording: Conductor's announcement says explicitly, "We have never sold or trained on Conductor user data, and we still won't. We'll delete all Conductor user data when the app shuts down." At a time when AI tools routinely feed on user data, that sentence is a statement in itself. Whether it holds up is something to check after November 20, and again when the local app winds down in early 2027. It cannot be verified in advance — only after the fact.
VibeFix's take
This acquisition carries two warnings for people doing vibe coding. The first is about tool choice: the orchestration layer is becoming the new battleground. When picking agent tools, do not only ask which model it plugs into — ask how it manages multiple agents: how cleanly workspaces are isolated, how smoothly changes merge, how conflicts are handled. Conductor being bought is proof that this layer is valuable.
The second warning is more practical: a niche tool's lifespan may be shorter than you think. A product live for barely a year, fresh off a Series A, can still be shut down overnight. When you build a core workflow on a small team's product, keep an exit route: code in a Git repository, conversation records exported regularly, and never keep everything in a single cloud workspace.
One last reminder on dates: for Conductor Cloud users, November 13 is the real deadline. Transfer workspaces to local first, then export a full data archive. Do not wait until the 20th — that is already the day data deletion begins.
One more thing about where the team is going: it is part of the deal itself. Holtz wrote on X that he looks forward to working with Alexandr Wang and Nat Friedman, and called Superintelligence Labs the place with "the highest talent density of all the labs." That is partly ceremony, but the direction is clear: Conductor's people are not being absorbed to maintain an old product, but to build something new inside Meta. The independent product's life is over, but its team and ideas will live on in Meta's next interface. What users can do now is carry their data and workflows out safely, then wait and see what Meta serves up early next year.
A third warning is for independent developers: when choosing a tool, check whether it offers a dignified exit route before judging how good its features are. Conductor at least gave users two ways out — one-click transfer of workspaces to local, and a full data export — and many niche tools offer neither. No matter how smooth a tool feels, without export and a local fallback, you will have no moving van on the day it shuts down. Before adopting a small team's product next time, look for the export button in its settings first: if you can find it, then talk about how good it is; if you cannot, treat it as something that may disappear at any moment.
Sources: TechCrunch, "Meta acquires AI coding startup Conductor," November 4, 2026; Conductor's official announcement, "Conductor is joining Meta," and changelog 0.93.0, November 4, 2026. Deal terms were not disclosed.
Comments (0)
Sources
Related articles
On 18 November 2026, Google Research and co-authors posted BuggyMLE to arXiv: bugs injected into 222 working ML repositories, and 14 models slide from 95% on the easiest tier to 9% on the hardest. A 9-billion-parameter model trained easy-to-hard beats Claude Opus 5.5 on the 10-bug tier.
On 17 November 2026, Microsoft launched MAI Code 2 and Code 2 Flash: the larger model has 77 billion active parameters and a one-million-token context, averaging 63.2% across four benchmarks — close to GPT-6.1 Sol and Claude Opus 5.5 — at $0.309 per task. Read the three caveats first: promotional pricing, the lowest score of the three, and no long-horizon work.
On 11 November 2026, AWS open-sourced UAEF: adapters turn traces from LangGraph, LangChain, Strands and Bedrock AgentCore into one format, then 36 metrics across 7 dimensions score them, flagging regressions past 5%. The judge model is hardened too — scores must cite verbatim evidence, and never decide safety.