Here's a secret about AI agents: the impressive part isn't the LLM. It's what happens when you give the LLM permission to call your code.

A method that looks up customer data. A function that checks if inheritance shares add up to 100%. A service that retrieves FAQ content. You've been writing AI tools for years. They just didn't have a [Description] attribute on them yet.

In this session, I'll start with the simplest possible AI agent, then progressively add real capabilities: function tools backed by plain C# methods, middleware for logging and control, a multi-agent workflow where one agent validates legal documents while another drafts summaries, and full observability through OpenTelemetry. The examples come from our legal tech product, where we went through exactly this journey, from auto-suggesting chat answers to orchestrating agents that review legal documents.

Built on Microsoft Agent Framework and Microsoft.Extensions.AI, but the talk is about the patterns, not the packages. If you know dependency injection, middleware, and handlers, you're already 80% of the way there.

This session is for .NET developers who want to integrate AI capabilities into existing applications without rewriting everything or getting a PhD first.

You'll leave with:
- A practical understanding of agents, tools, and workflows in Microsoft Agent Framework
- The ability to decide what should be a deterministic function vs. what needs an LLM
- Patterns for multi-agent orchestration that are testable and observable
- A healthy respect for the line "if you can write a function, do that instead"