How to Enable Agent Teams in Claude Code (Experimental Feature)

Feb 9, 2026

Purpose

This post demonstrates how to enable Agent Teams in Claude Code, an experimental feature that allows multiple AI agents to work in parallel on complex tasks.

Environment

Claude Code (Opus 4.6+)
settings.json configuration file
Operating System: Any (macOS, Linux, Windows)

The Problem

When I give Claude Code a complex refactoring task, it works through everything sequentially. Even with fast AI responses, large tasks take time because one agent handles everything alone.

A developer on Reddit reported that Claude Code spawned three specialized agents for a refactoring task:

One agent worked on backend
One agent worked on frontend
One agent played code reviewer

They completed the work in 15 minutes. The terminal split into 3 panes showing all three agents working simultaneously. The agents communicated with each other, challenged approaches, and coordinated independently.

I wanted to try this feature.

How to Enable Agent Teams

First, I need to find my Claude Code settings.json file.

Here’s the default location:

# macOS
~/Library/Application Support/Claude Code/settings.json

# Linux
~/.config/Claude Code/settings.json

# Windows
%APPDATA%/Claude Code/settings.json

Now I read the current settings:

{
  "env": {}
}

The env section is empty. I need to add the experimental flag:

{
  "env": {
    "CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS": "1"
  }
}

I save the file and restart Claude Code.

What Happens Next?

Now when I give Claude a complex task, it automatically decides whether to spawn multiple agents.

Here’s an example prompt:

> "Refactor the user authentication module. Split validation
  logic into a separate service. Update the API endpoints
  to use the new service. Add error handling and logging."

Claude spawns three agents:

# Terminal Pane 1: Backend Agent
[Backend]: Analyzing authentication module structure...
[Backend]: Extracting validation logic to UserServiceValidator
[Backend]: Implementing validateCredentials() method

# Terminal Pane 2: Frontend Agent
[Frontend]: Reviewing auth component integration
[Frontend]: Updating API calls to match new endpoint structure
[Frontend]: Adding loading states for validation

# Terminal Pane 3: Reviewer Agent
[Reviewer]: Backend, ensure validation errors return proper status codes
[Backend]: Updating to return 400 for validation failures
[Reviewer]: Frontend, the error handling needs to match new API response format
[Frontend]: Adjusting error parsing logic

The agents message each other directly. They coordinate without my intervention.

Why This Matters

I can think of several benefits:

Parallel Processing: Three agents working simultaneously finish faster than one agent working sequentially
Specialization: Each agent focuses on their expertise (backend, frontend, review)
Quality: The reviewer agent catches issues that backend/frontend agents might miss
Communication: Agents debate approaches and challenge each other’s decisions

The Reddit developer summed it well:

“I’ve coded for 6 years. First time I’ve genuinely felt like my job is shifting from ‘writes code’ to ‘directs AI team that writes code.’”

Common Mistakes

I tried this with a simple task first:

> "Add a console.log statement to this function"

Claude didn’t spawn agents. It just made the change directly.

I think the key is complexity. Agent Teams activate for:

Multi-file refactoring
Architecture changes
Feature implementation across backend/frontend
Tasks requiring different expertise

Simple edits don’t trigger it.

Summary

In this post, I showed how to enable Agent Teams in Claude Code. The key point is adding the CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS environment variable to settings.json. Once enabled, Claude can spawn multiple specialized agents that communicate and work in parallel on complex tasks.

Final Words + More Resources

My intention with this article was to help others share my knowledge and experience. If you want to contact me, you can contact by email: Email me

Here are also the most important links from this article along with some further resources that will help you in this scope:

👨‍💻 Reddit Discussion - Claude Code Agent Teams
👨‍💻 Claude Code Documentation

Oh, and if you found these resources useful, don’t forget to support me by starring the repo on GitHub!