Agent teams enhance productivity by allowing multiple teammates to work on tasks in parallel. They’re particularly effective in scenarios like research, feature development, debugging, and coordinating changes across different layers of a system. Each member can focus on a distinct aspect of a problem, which prevents overlap and accelerates the discovery process. For example, in debugging, teammates can explore different hypotheses simultaneously, leading to quicker resolutions. Unlike subagents, which report results back to a main agent, agent teams facilitate direct communication among members. Each team member operates independently, but they share a task list for coordination. This setup is ideal for complex projects that require discussion and collaborative input. While subagents are suited for focused tasks where the result is the only concern, agent teams shine in environments where conversation and collaboration are necessary. To get started with agent teams, users must enable them through an environment variable. Once active, you can create a team by instructing Claude in natural language. The lead agent coordinates tasks and can control teammates, including assigning tasks or allowing them to self-claim. Two display modes are available: in-process, where all teammates operate within the same terminal, and split panes, which provides a more visual layout. Task management is straightforward, with each task being marked as pending, in progress, or completed. The lead can assign tasks or teammates can self-claim them once they finish their current work. Notifications keep everyone informed about task status, and hooks can enforce quality checks on completed tasks. The structure is designed for efficiency, enabling teams to adapt quickly to the demands of their projects.