Picture a Formula 1 car. The engine is extraordinary — but without the chassis, suspension, telemetry, and pit crew systems around it, it doesn't race. It just sits there.
Your AI agent has the same problem. You've picked a great model. But what's actually running it?

This session draws a sharp line between AI frameworks (which build agents) and agent harnesses (which operate them) — then shows exactly where Microsoft Foundry fits. We build a production-ready agentic workflow live on stage: model invocation, tool chaining, state persistence, error recovery, and real-time observability — all inside Microsoft Foundry.
The use case is concrete: a multi-step internal agent that coordinates data retrieval, applies business rules, and handles partial failures without falling over. You'll see the exact patterns that separate a polished demo from a production system.

Whether you're starting your agentic journey or hitting limits with your current setup, this session gives you a clear architectural mental model and hands-on patterns you can take directly into your Azure environment.
By the end, you'll be able to:

- Explain the harness pattern and why it matters for production agents
- Use Microsoft Foundry to orchestrate, observe, and deploy agentic workloads
- Design agents that handle real-world failure modes