Build, Trace, Test: Production-Grade AI Agents

Build, Trace, Test: Production-Grade AI Agents with .NET Microsoft Agent Framework

Building an AI agent that works in a demo is surprisingly easy. Building one that you can change, debug, evaluate, and trust is a much more interesting engineering problem.
In this session, we'll build a real agent in C# using Microsoft Agent Framework and connect it to organizational context from Microsoft 365 using Work IQ. Then we'll go beyond the happy-path demo and treat the agent like production software.
We'll instrument agent execution with OpenTelemetry to see model calls, tool execution, latency, token usage, and failures. We'll create an evaluation harness that turns expected agent behavior into repeatable tests, combining deterministic checks with model-based evaluation. Finally, we'll intentionally change the agent and use our telemetry and evaluations to determine whether we actually made it better—or quietly introduced a regression.

Along the way, we'll explore the evolving Microsoft agent development stack, including Agent Framework, Work IQ, A2A, OpenTelemetry, and Microsoft.Extensions.AI evaluation.
If you're a .NET developer trying to understand what engineering AI agents looks like after the first successful prompt, this session is for you.

Speaker: Brian Haydin