Every capability you give your AI agent can make it even better, but different tools have different strengths and weaknesses. Between RAG, MCP, tools, skills, CLIs, and APIs, how do you know which one to choose?

In this session, Spencer will break down the extensibility architecture that makes production AI agents actually useful. We'll go over how MCP gives you the connection layer to external systems, RAG gives you the knowledge layer for domain-specific retrieval, and skills give you the behavior layer for complex multi-step workflows - and more importantly, how to compose them in real production systems. We'll cover the practical decisions: when to use MCP vs. direct tool calls, when RAG beats stuffing context into the prompt, when you should go basic and shell out to a CLI, and how to test all of it.

Come with an AI agent that can answer questions. Leave knowing how to make it actually do things.