Claude Code Agent Teams
Master parallel AI collaboration with agent teams. Learn to coordinate multiple Claude instances for complex debugging, code reviews, and cross-layer development.
Why take this course?
Unlock the power of parallel AI collaboration. When production is down and you need answers in minutes, not hours, agent teams let you investigate multiple hypotheses simultaneously. Learn to coordinate multiple Claude instances, implement competing hypotheses debugging, and orchestrate complex cross-layer development. From setup to advanced patterns, this course gives you the skills to leverage team-based AI for research, review, and development at scale.
Prerequisites
This course builds on concepts from the following courses. It is recommended to complete them first:
Course Modules
Learning Goals
- Understand when and why to use agent teams vs single sessions or subagents
- Identify tasks that benefit from parallel exploration
- Distinguish between subagents and agent teams
Concept Card Preview
Visuals, diagrams, and micro-interactions you'll see in this module.
Sarah's Friday Deadline
It's 2 PM on Friday. Sarah's team lead announces: Production is down.
The error could be anywhere — the frontend Re…
The Sequential Investigation Trap
Sarah's problem isn't unique. When bugs strike, we face anchoring bias — the tendency to fixate on the first plausib…
Enter Agent Teams: Parallel Minds
Sarah discovers Agent Teams — a way to coordinate multiple Claude Code instances working together.
She spawns four…
Learning Goals
- Understand the three-layer architecture of agent teams
- Learn how teammates share information through the mailbox system
- Know where team configuration and tasks are stored
- Understand context boundaries between lead and teammates
Concept Card Preview
Visuals, diagrams, and micro-interactions you'll see in this module.
The Three Components of Every Team
An Agent Team has three distinct components. The Team Lead is your main Claude session — you create the team, spawn…
How the Task List Coordinates Work
The shared task list is the heartbeat of coordination. Tasks move through three states: Pending (created, not starte…
The Mailbox: Direct Communication
Unlike subagents that only report to the caller, teammates can message each other directly through the mailbox.
**Direc…
Learning Goals
- Configure display modes (in-process vs split panes)
- Create your first agent team with proper prompts
- Navigate between teammates and communicate effectively
- Understand task claiming and assignment
Concept Card Preview
Visuals, diagrams, and micro-interactions you'll see in this module.
Enabling Agent Teams
Agent Teams are experimental and disabled by default. Enable them with an environment variable:
`export CLAUDE_CODE…
Creating Your First Agent Team
Creating a team is natural language: describe what you want, and Claude evaluates whether it warrants parallel inves…
Choosing Your Display Mode
Two ways to view teammates:
In-process mode (default): all teammates run in your terminal. Navigate with **Shift+Up…
Learning Goals
- Master the competing hypotheses pattern for debugging
- Implement plan approval workflows for high-risk changes
- Use delegate mode for pure orchestration
- Apply quality gates with hooks
Concept Card Preview
Visuals, diagrams, and micro-interactions you'll see in this module.
The Competing Hypotheses Pattern
A single investigator tends to find one plausible explanation and stop looking. The competing hypotheses pattern fig…
Plan Approval: Quality Gates for Complex Work
For risky tasks, require teammates to get approval before implementing. The workflow: teammate works in **read-only plan…
Delegate Mode: Pure Orchestration
By default, the lead might start implementing instead of waiting for teammates. Delegate mode restricts the lead to…
Learning Goals
- Understand token cost implications of agent teams
- Know the hard limitations and how to work around them
- Apply battle-tested best practices
- Properly clean up teams when finished
Concept Card Preview
Visuals, diagrams, and micro-interactions you'll see in this module.
Understanding Token Costs
Each teammate is a full Claude instance with its own context window. Token costs scale linearly: 3 teammates ≈ 3x th…
Hard Limitations to Know
Agent Teams are experimental. Key limitations:
No session resumption: /resume doesn't restore teammates. After re…
Battle-Tested Best Practices
Give teammates enough context: they get CLAUDE.md but NOT your conversation. Include task-specific details in spawn…