Meta, OpenAI and GitHub bet on multi-agent coding

In one week, three of the biggest companies in software bought or shipped tools for the same idea: running several AI coding agents at once. Meta acquired Conductor, a Mac app that runs Claude Code, Codex and Cursor agents side by side. OpenAI agreed to buy Pierre, whose Git infrastructure is built for agents. And GitHub put Copilot Projects, which fans one goal out to parallel agents and hands back pull requests, into public preview. Smaller launches from Kiro, Tessl and LaunchDarkly landed in the same days.

Here is what each one is, what changes if you use Conductor, and what the pattern says about where AI-assisted development is going.

What happened

Meta buys Conductor (4 November)

Conductor describes itself as a way to "run parallel coding agents on your Mac": it creates isolated copies of your codebase, runs an agent in each, and lets you review and merge what they produce. It supported Claude Code, Codex and Cursor agents, with users paying for those agents through their own subscriptions and API keys. Founded in 2024 by Charlie Holtz and Jackson de Campos, it raised a $22 million Series A in March 2026 led by Matrix and Spark.

The deal was reported by Axios and confirmed the same day by Conductor ("Conductor is joining Meta") and by Meta, which declined to give TechCrunch any further details. No price was disclosed. Conductor's own post says the team has "big, crazy plans for the next interface" and will have "something new for you to use early next year."

OpenAI to acquire Pierre (4 November)

Pierre Computer Company makes Code Storage, "Git infrastructure at scale" built for AI and agent platforms, whose listed customers include Lovable and Bolt. It also makes open-source developer tools, and OpenAI says anyone who has browsed files or reviewed a change in the Codex app has already used Pierre's Trees and Diffs libraries.

OpenAI's stated aim, as reported by InfoWorld, is to bring that Git infrastructure into Codex, where it has been working on Git's limits when humans and agents work in parallel. Code Storage customers can keep using it after the deal closes while its features move into Codex, and Pierre's open-source projects stay open. The deal still needs customary approvals, including regulatory ones, and no price was given.

GitHub ships Copilot Projects (28 October)

At its Universe conference, GitHub relaunched the Copilot app as "designed from the ground up for agentic software delivery." The headline feature is Projects, now in public preview: you describe a goal, attach repositories, design mocks or Slack threads, and Copilot "run[s] agent work across multiple repos in fully managed, parallel sessions," returning pull requests for review. GitHub has not said when Projects will be generally available or which plans include it. The Copilot app itself has been available on macOS, Windows and Linux, on every Copilot plan including Free, since July.

And the same week

• Kiro bots (5 November): named, long-running agents in Kiro Crew, each with its own memory, sessions and scoped permissions, able to coordinate parallel worker sessions. Generally available.
• Tessl (4 November): an agent-agnostic platform for "software factories" that chains reusable skills into workflows such as ticket-to-PR and code review, and works with Claude Code, Cursor, Copilot and Gemini.
• LaunchDarkly Release Autopilot (announced 4 November, generally available 16 November): "automated release control for AI-built code" that rates each pull request's risk, wraps it in a feature flag, rolls it out gradually and rolls back automatically.

If you use Conductor: the dates that matter

Conductor's announcement sets a short timeline:

Date	What changes
4 November	No new sign-ups or downloads; free users can no longer upgrade to Pro
13 November	Conductor Cloud workspaces become read-only (you can still read chats and export)
20 November	Conductor Cloud shuts down and user data deletion begins
Early 2027	The local app stops; until then it gets new models and critical fixes

Pro subscribers keep Pro features for free, and active subscribers get a refund. If you used Conductor Cloud, use its "Transfer to local" option before 20 November, or export what you need. Conductor says it will delete all user data when the app shuts down and has "never sold or trained on Conductor user data."

You do not need a product to keep the core workflow. Conductor's central trick, one isolated copy of the repository per agent, is what Git's worktrees already do:

git worktree add ../auth-refactor -b auth-refactor
git worktree add ../billing-tests -b billing-tests
# run one agent in each folder; each works on its own branch
git worktree list
git worktree remove ../auth-refactor   # after merging

Each agent gets its own folder and branch, so two agents never edit the same checkout, and you review each branch as a normal pull request.

What it means for developers

The bottleneck has moved from writing code to reviewing it

Every one of these launches is about what happens around the agents: isolating their work, coordinating them, reviewing what they return and releasing it safely. None of them is a better model.

The numbers in the coverage point the same way. LaunchDarkly's own survey found 94% of respondents saw faster coding from AI, and 91% had become more cautious about pushing to production. GitHub prices review by effort in AI credits; InfoWorld reports averages of about $0.25 for a light review and $10 for the deepest. LaunchDarkly's product lead put it plainly: "Human attention is among the scarcest resources in the software factory."

If you run agents in parallel, your review queue grows faster than your output. Plan for that: small, focused tasks per agent, tests that run on every branch, and feature flags for anything risky.

Independent tools are being absorbed

Cursor became part of SpaceX in August. Now Conductor goes to Meta and Pierre to OpenAI. The Decoder described the pattern as labs buying tooling startups to "lock developers into their own platforms."

Conductor's appeal was that it sat above the vendors: one window for Claude Code, Codex and Cursor. That kind of neutral layer is exactly what is disappearing. The practical defence is to keep your workflow in open, portable pieces you own:

• Instructions in one AGENTS.md that any agent can read, instead of rules locked into one tool. I set this up across Cursor and Claude Code earlier this year; the Cursor vs Claude Code post shows how.
• Plain Git for isolation and review: worktrees, branches and pull requests work with every agent and outlive every app.
• Your own subscriptions and keys, so switching the tool on top does not mean losing access to the models underneath.

Git is becoming agent infrastructure

OpenAI buying a Git infrastructure company is the least flashy deal of the week and maybe the most telling. Agents create branches, commits and diffs far faster than people do, and OpenAI says it has been working on Git's limits for exactly that case. The plumbing built for human pace is being rebuilt for agents, and your Git host, diff viewer and review tools are where you will notice it.

The short version

• Conductor users: move anything in Conductor Cloud before 20 November; the local app runs until early 2027.
• Copilot users: Projects is GitHub's answer to running agents in parallel, built into the app you may already have, but it is a public preview with no published pricing or plan list yet.
• Everyone else: you can run agents in parallel today with Git worktrees and the agent you already pay for. The scarce resource is your review time, so spend money and effort there before adding more agents.

Frequently asked questions

What happens to Conductor now that Meta has bought it?

New sign-ups and downloads stopped on 4 November 2026. Conductor Cloud workspaces go read-only on 13 November and the service shuts down on 20 November, when user data deletion starts. The local Mac app keeps working, with new models and critical fixes, until early 2027. Pro features are free for existing subscribers, and active subscribers get a refund.

Is GitHub Copilot Projects available?

Yes, in public preview since 28 October 2026, inside the GitHub Copilot app. You describe a goal, add repositories and context, and Copilot runs several agents in parallel and returns pull requests. GitHub has not published a general-availability date or which plans include it.

How do I run several coding agents in parallel without Conductor?

The core idea is an isolated copy of the repository per agent. Git's own worktrees do that: git worktree add ../feature-x -b feature-x gives each agent its own folder and branch, so their changes never collide, and you review and merge each branch as usual.