The Anatomy of an AI Agent

Terms such as model, instructions, orchestration, tools, context, memory, retrieval, and MCP are often used when describing AI agents, but it isn’t always clear what role each one plays or how they fit together. This session takes an AI agent apart and follows a request as it moves through the system.

You’ll see how the model, instructions, orchestration, tools, context and knowledge, and state or memory work together, and what distinguishes an agent from a simple call to a language model. We’ll look at how tools let an agent interact with external systems, where MCP fits into that integration, and what MCP does and doesn’t provide. Along the way, we’ll examine the roles of identity, authorization, human approval, and observability, and discuss where model-driven decisions are useful versus where conventional application logic is still the better choice.

By the end of the session, you’ll have a practical mental model for the components that make up an AI agent and the responsibilities that still belong to the application around it.