Skip to main content

Claude Code Now Runs Multiple AI Threads at Once

Anthropic Turns Claude Code into a Multi-Agent Team

Anthropic has rebuilt the core of Claude Code, adding a multi-threaded parallel collaboration system. Instead of one AI assistant grinding through a task alone, you now describe your project goal, and a coordinator breaks it into pieces. Each piece goes to an independent "thread" running in its own cloud session.

Think of it as assembling a small team of developers—except they're all AI, and they never sleep.

How It Works

You start by describing what you want to build. The coordinator analyzes the goal and splits it into parallel tasks. Each thread then works on its own slice, running in a separate cloud session. You can watch the progress in the main chat room or dive into any individual thread to see what's happening. Mobile support is included, so you can check in from your phone.

But these threads aren't isolated. Claude continuously builds shared memory between them. That means each thread knows what the others are doing—no duplicated effort, no conflicting changes. There's also a project library that collects user-uploaded files and results from every thread, weaving everything into one collaborative workflow.

Each thread can independently submit pull requests and run tests. If a thread hits a snag, it can adjust without waiting for the others. It's a far cry from the single-threaded back-and-forth most AI coding tools still use.

Who Can Try It—and What Comes Next

For now, the cloud session feature is in testing for some Pro and Max subscribers. Team and enterprise versions are on the roadmap. Local execution is also in development, which could appeal to developers who want to keep their code on their own machines. You can apply for access through a waiting list.

This update pushes Claude Code beyond a solo coding assistant. It's now a platform where multiple AI agents collaborate on complex development tasks. Earlier, Anthropic made the autonomous driving mode of Claude Code the default and claimed its performance on some security tasks beats human developers.

As more work shifts to autonomous agents running in parallel, three questions loom large: computing resource consumption, cost of use, and how developers control agent behavior. Those are the new pressure points for anyone building with AI coding tools.

Key Points

  • Multi-threaded parallel execution: A coordinator splits your project goal into independent threads, each running in its own cloud session.
  • Real-time tracking: Monitor progress from the main chat or individual threads, including on mobile.
  • Shared memory & project library: Threads share context and results, enabling true collaboration.
  • Autonomous pull requests & tests: Each thread can submit PRs and run tests on its own.
  • Limited availability: Testing for some Pro and Max users; team, enterprise, and local versions coming later.