Meta has acquired Conductor, a startup whose Mac app lets developers run several AI coding agents at the same time, each working in its own isolated copy of a codebase. Axios first reported the deal on November 4, 2026, Conductor confirmed it on its blog, and terms were not disclosed. The team joins Meta Superintelligence Labs. For everyone else, the acquisition is a signal: parallel agents, not one agent at a time, is where coding work is heading.
What Conductor does
Conductor’s pitch is simple. Instead of typing a task into one agent and waiting, you open several lanes, hand each one a task, and let them work at once. Each agent gets its own git worktree, so a change in one lane cannot break another. When an agent finishes, you review the diff and merge, or send it back with notes.
The app is agent-agnostic. It supports Claude Code, OpenAI Codex, Cursor’s agent, and OpenCode, according to SiliconANGLE. Conductor’s website claims more than 100,000 users. The company raised a $22 million Series A in March 2026 from Spark Capital and Matrix.
What Meta gets
Meta is buying a team and a workflow, not a model. CEO Charlie Holtz and CTO Jackson de Campos join Meta Superintelligence Labs (MSL), where they will work with Nat Friedman and Alexandr Wang, per TechCrunch. Friedman ran GitHub when Copilot launched, so the interest in developer tooling is not new.
The likely read is that Meta wants orchestration know-how for its internal engineering and for whatever coding product MSL ships next. Conductor had spent two years learning how developers actually manage several agents: how to show progress, when to interrupt, how to keep merges sane. That is hard-won product knowledge, and it is cheaper to buy than to rediscover.
What happens to current users
Holtz said core features keep working “until early next year.” The one hard date is November 20, 2026, when Conductor’s cloud service shuts down. After that, cloud data is deleted.
If you use Conductor, do these before November 20:
- Export any conversation history, task logs, or settings you want to keep.
- Make sure every worktree you care about has been merged or pushed to a remote. Local worktrees are plain git; they are not affected by the cloud shutdown, but check anyway.
- Pick a fallback. The section below covers how to get most of the value with git alone.
Why parallel agents are becoming the default
A single coding agent is bounded by one thing: you have to wait for it. A task that takes an agent twelve minutes leaves you idle for twelve minutes, or context-switching in a way that makes the review worse.
Parallel lanes change the math. Three agents on three scoped tasks finish in roughly the time of one, and your job shifts from typing prompts to reviewing output. That is a better use of a senior engineer’s attention, and it is why the pattern has spread from a few power users to a product category.
The adoption data backs this up. The Stack Overflow 2026 Developer Survey, with 30,903 responses, found that among developers using coding agents, 66% use Claude Code and 59% use GitHub Copilot. Automated agent workflows are used by 26.2% of respondents, per the survey results post. A quarter of developers already running agents on autopilot is the audience Conductor was built for.
The same week, elsewhere
Meta is not the only one betting on agents that run without a human watching. AWS launched Kiro bots, agentic bots with memory for recurring development jobs such as dependency updates and test fixes, per VentureBeat. On Product Hunt, a tool called AgentTM reportedly offers time-travel debugging for coding agents, letting you step back through an agent’s actions. Both point the same way: more agents, longer sessions, less direct supervision.
How to run parallel coding agents yourself
You do not need Conductor to do what Conductor did. The core idea is one git worktree per task, one agent per worktree, and a review process that keeps you honest.
Step 1: one worktree per task
git worktree add ../repo-auth-fix -b agent/auth-fix
git worktree add ../repo-pagination -b agent/pagination
git worktree add ../repo-flaky-tests -b agent/flaky-testsEach directory is a full checkout on its own branch, sharing the same .git store. Agents cannot step on each other’s files.
Step 2: one agent per worktree
Open a terminal per worktree and start your agent of choice in each. Claude Code, Codex, OpenCode, and Cursor’s agent all work from a directory, so this is just cd and launch. Give each agent a brief that fits in a paragraph and has a clear done state.
Step 3: keep tasks small and well scoped
Parallel work fails when tasks overlap. Good lanes: fix a bug in one module, add a migration, write tests for an existing function. Bad lanes: “refactor the API” and “update the API docs” at the same time. If two tasks touch the same files, run them in sequence.
Step 4: a review queue
Treat finished worktrees as a queue, not a pile. Review one at a time, in the order they finish, and do not merge anything you have not read. A short checklist helps: does the diff match the brief, are there tests, did it touch files outside its scope?
Step 5: CI as the arbiter
Push every agent branch and let CI run before you review. Green builds get your attention first; red builds go back to the agent with the failing log pasted in. This stops you from spending review time on work that does not compile.
| Conductor | DIY with git worktrees | |
|---|---|---|
| Isolation per agent | Worktree, managed | Worktree, manual |
| Supported agents | Claude Code, Codex, Cursor, OpenCode | Anything that runs in a directory |
| Progress view | Single UI | Several terminals or tmux |
| Cloud sync | Until Nov 20, 2026 | None needed |
| Cost | Was free to start | Free |
Our take
Conductor proved the workflow, and Meta decided it was worth owning. That is a loss for the independent tools market and a validation of the pattern. The good news is that the pattern itself is just git. Worktrees have been in git since 2015, and every major coding agent runs from a plain directory.
If you are new to agents, start with our primer on what an AI agent is, then run two lanes on two small tasks and see how the review feels. If you are already running several agents, the lesson from this week is to keep your setup portable. Tools get acquired. Branches do not.
We track the coding agent landscape in our tools directory and send the week’s agent news every Monday in the newsletter.
Questions, answered
What is Conductor and what did Meta buy?
Conductor is a Mac app that runs several coding agents in parallel, each in its own copy of a codebase, so a developer can review and merge their work. Meta acquired the company on November 4, 2026, for an undisclosed sum, and the team is joining Meta Superintelligence Labs.
Will Conductor keep working after the Meta acquisition?
For now. CEO Charlie Holtz said core features will keep working until early next year. The cloud service shuts down on November 20, 2026, and cloud data is deleted after that date, so export anything you need before then.
How do I run multiple coding agents in parallel without Conductor?
Create a git worktree per task, start one agent in each worktree with a small, well-scoped brief, and let CI decide which branches are ready to merge. It takes a few shell commands and no extra software.
