Orchestrating Agent Teams in Claude Code

Agent teams take delegation to the next level — instead of individual subagents that report back in isolation, you get multiple Claude Code sessions working simultaneously on a shared task list with built-in coordination. In this lesson, Brian walks through the full agent teams workflow, from enabling the experimental feature flag in settings to prompting Claude to spawn a three-agent team (backend, frontend, and tester) that collaborates on building a project dashboard feature for the demo app.

Brian shows the agent team in action — how the lead agent creates tasks with dependencies, how teammates self-assign work, and how to navigate between agents in the terminal UI. He also covers the practical realities: agent teams burn through tokens faster since each teammate is a full Claude Code session, the UI for switching between agents takes some getting used to, and you should think of your prompt like writing tickets for a real engineering team. The lesson wraps up with a clear comparison of when to reach for subagents versus agent teams.

• Enabling the experimental agent teams flag in settings.json
• Prompting Claude to create an agent team with specific roles and responsibilities
• How the team lead orchestrates task creation with dependencies
• Navigating between teammate views in the default terminal UI
• Live demo: three agents building a dashboard feature in parallel
• Debugging when agent team output has issues (screenshot workflow)
• Token cost reality: 3-4x usage with multiple agents
• Shared task list with dependency tracking and state management
• The mailbox system for inter-agent communication
• Decision framework: subagents (independent work) vs. agent teams (coordinated work)