Agent Builder

This teaches you to build AI agents by getting out of your own way. The core insight is that Claude already knows how to be an agent, you just need to provide capabilities (what it can do), knowledge (what it knows), and clean context. Start with 3-5 capabilities in a simple loop where the model decides whether to act or respond. Most agents never need subagents or complex planning. The philosophy is refreshingly anti-engineering: trust the model to reason, don't pre-specify workflows, add complexity only when real usage demands it. Includes minimal working examples and templates for capabilities.