How and when to use subagents in Claude Code

Claude Code supports subagents — isolated Claude instances with their own context windows — to handle parallel, research-heavy, or independent tasks without polluting the main session. The post covers when to use subagents (parallel work, codebase research, fresh code reviews, pipeline workflows), how to invoke them (conversational prompts, custom subagent definitions in .claude/agents/, CLAUDE.md policies, reusable skills, and lifecycle hooks), and when to avoid them (sequential dependent work, same-file edits, small tasks, or tasks requiring inter-agent communication). Practical prompt patterns and configuration examples are provided for each approach, along with guidance on when to use agent teams instead.