In this video I’m looking at Claude Code sub-agents and background agents, because understanding this is a critical skill if you’re vibe coding or doing agentic coding as a serious dev. The big idea is simple: stop trying to do everything in the main thread. Your main conversation is your most valuable resource, because once you chew through the context window, quality drops, compacting kicks in, and you lose important details. So I show you how to list the built-in agents (Explorer, Planning, Bash, etc.), invoke them directly with @, and run them in the background with Ctrl+B so you can keep moving while they work. Then I walk through why sub-agents are so powerful: they run in their own threads with their own token budgets, and they only return a summary back to the main agent. That doesn’t necessarily reduce total tokens, but it protects your main context so you don’t hit the wall on bigger builds. I also show how to create your own specialized agents with /agents—like a UI expert with a neo-brutalism design system, a high-quality coder agent, and a lightweight code reviewer. Finally, I share my practical workflow: plan in parallel (multiple planning agents), write specs to files, then orchestrate “waves” of implementation where coder agents build in parallel and reviewer agents validate, iterate, and do final multi-perspective reviews. That’s how you ship complex apps—like a full kanban todo app with optional auth, Postgres, and Drizzle—without your main thread turning into a dumpster fire.