The era of a single AI is over. Time to build a team.
16 AI agents worked simultaneously to build a 100,000-line C compiler. Cost: $20,000. Timeline: 2 weeks. Lines of code written by humans: zero.
What Is It?
Honestly, this one gave me chills when I first saw it.
On February 5, 2026, Anthropic announced Claude Opus 4.6 alongside a feature called Agent Teams. In simple terms, it lets you run Claude Code — Anthropic's AI coding assistant — as multiple agents simultaneously. And they can talk to each other.
You could already have AI write code for you. But it was one task at a time, sequentially. "Do this" → wait → "Now do this" → wait. Agent Teams flips this entirely. A single team lead breaks down the work and distributes it to multiple team member agents, who work independently while exchanging messages with each other.
Google's Addy Osmani called this a "Swarm" and nailed the core insight: "LLMs degrade as context grows larger." So instead of dumping everything on one agent, you assign narrow scopes to multiple agents, improving each one's reasoning quality.
According to VentureBeat, Opus 4.6 launched with support for a 1-million-token context window alongside Agent Teams, and Anthropic's enterprise adoption rate skyrocketed from nearly 0% in March 2024 to about 40% by January 2026. Once you try it, you'll understand why.
What's Different?
It goes from "errand runner" to "actual team."
Claude Code already had "subagents." The main AI would summon helper AIs, assign tasks, and receive results. But subagents couldn't talk to each other. They worked in isolation and only reported back.
Agent Teams is a completely different game.
| Subagents (Previous) | Agent Teams (New) | |
|---|---|---|
| Communication | Report only to main AI | Team members talk directly to each other |
| Coordination | Main AI manages everything | Self-coordinate via shared task list |
| Best for | Simple tasks where only results matter | Complex tasks requiring discussion and collaboration |
| Cost | Relatively cheaper | Scales proportionally with team size |
Anthropic ran their own experiment to prove this — that's the C compiler project mentioned above. They placed 16 agents in individual Docker containers with a file-locking system for coordination, and out came a 100,000-line compiler capable of building Linux kernel 6.9. 99% pass rate on the GCC test suite. Successfully compiled PostgreSQL, Redis, and FFmpeg.
The developer community is buzzing too. People are reporting things like "I ran PR reviews from three perspectives simultaneously — security, performance, and test coverage — and caught things I'd missed reviewing solo" and "When debugging, I have five hypotheses tested in parallel, which eliminates confirmation bias".
Pricing? The Max 5x plan is $100/month, Max 20x is $200/month. You'll need at least Max 5x to properly run Agent Teams, and Max 20x is recommended for serious use. Pay-as-you-go API pricing ($5/million input tokens, $25/million output tokens) is also available.
Quick Start Guide
- Install Claude Code
npm install -g @anthropic-ai/claude-code claude # Authenticates on first run - Enable Agent Teams
// Add to settings.json { "env": { "CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS": "1" } } - Create a team with natural language
Create an agent team to review this project. - 1 security expert - 1 performance analyst - 1 test verification lead Have each review independently and share findings. - Coordinate your team
Shift+Up/Down: Switch between members /Ctrl+T: View task list /Shift+Tab: Delegation mode - Review results and wrap up
Say "wrap up the team" and you're done. Start with safe, read-heavy tasks like code reviews or research.
That's it. Start with safe, read-heavy tasks like code reviews or research. Once you try it, you won't go back to working solo.
Want to Go Deeper?
Anthropic Official Docs — Agent Teams
The most accurate reference for setup, usage, and limitations. Read this before getting started.
Anthropic Engineering Blog — Building a C Compiler
The full story of building a 100,000-line compiler with 16 agents. The best showcase of what Agent Teams can actually do.
Addy Osmani — Claude Code Swarms
A deep analysis of Agent Teams' architecture and mechanics by Google Chrome's engineering manager. Clear criteria for when to use them and when not to.
TechCrunch — Opus 4.6 Launch Coverage
Good for understanding the bigger picture of where Agent Teams fits in the AI industry.
https://techcrunch.com/2026/02/05/anthropic-releases-opus-4-6-with-new-agent-teams/
Claude Pricing Page
Compare Max plans and check Claude Code inclusion. Useful when deciding which plan fits your needs.
SitePoint — Agent Teams Setup Guide
A hands-on tutorial with screenshots walking you through setup step by step.
https://www.sitepoint.com/anthropic-claude-code-agent-teams/
