Stop Using Claude Code Like This (Use Sub-Agents Instead)

Description: In this tutorial the presenter explains sub-agents and background agents within Claude Code, showing how to create, configure and run specialized agents to analyze codebases, implement UI and backend tasks, and orchestrate parallel workflows. The video demonstrates building custom agents (UI expert, coder, code reviewer), planning complex features with multiple planning agents, and using background tasks to preserve the main context window while scaling implementation, testing and review across parallel coding and review agents. It also covers practical CLI commands and a full implementation example.

– Demonstrates built-in and custom sub-agents (bash, explore, planning, UI expert, coder, code reviewer), how to invoke them directly or via the main agent, and running them in the background to free the main thread.

– Explains context windows, token usage, and how sub-agents run in separate threads to prevent context compaction, improving quality on large codebases.

– Walks through creating agents, configuring models/tools, and applying a workflow: planning phase with multiple planning agents, splitting work into parallel tracks, coding agents implement tasks, and review agents iterate until complete.

– Shows a practical end-to-end example (Next.js app, Postgres/Drizzle, auth) where multiple background agents generate plans, produce files in parallel, fix issues, and reach a working application while keeping the main conversation compact.

Quotes:

Run any agent in the background with Ctrl+B to free the main thread.

Sub agents protect the main conversation from context compaction.

You can implement a whole app in parallel with coding and review agents.