Most people use Claude Code like a single brain. That’s the mistake. Subagents let you split work into clean, specialized units so you get speed without losing accuracy. The mental model: • You are the orchestrator • Subagents are contractors • Main context stays clean • Work runs in parallel when possible Rules that actually matter: • Use Explore for all codebase questions • Parallel execution for independent tasks • Sequential execution when order matters • One message = parallel, multiple messages = sequential • Background agents for slow tasks If you’re not orchestrating subagents intentionally, you’re leaving Claude Code’s biggest advantage on the table. This playbook teaches how to think, not just which buttons to press.