AI agents are no longer a research curiosity — they're shipping to production today. But most teams hit the same wall: a promising prototype that collapses under real-world complexity. Orchestrating multiple agents, managing state, handling failures gracefully, and integrating with enterprise systems turns out to be far harder than the demo suggested.

In this session, we go beyond the "hello world" agent and build something real. Using the Microsoft Agent Framework and .NET Aspire, we design a distributed multi-agent system that tackles a genuine enterprise modernization scenario — live, on stage, with code you can take home.

You will see how to compose specialized agents into reliable pipelines, how to coordinate agent collaboration and state across a distributed system, and how Aspire's orchestration and observability layer makes deploying and monitoring multi-agent workflows dramatically simpler. We cover the patterns that actually hold up in production: structured handoffs, tool grounding, observability with OpenTelemetry, and graceful degradation.

🤖 Understand the architecture behind production multi-agent systems with Microsoft Agent Framework
🔧 Design and wire specialized agents with well-defined roles, tools, and handoff contracts
🔗 Integrate agents with enterprise data sources and Azure services
⚙️ Use .NET Aspire to orchestrate, observe, and deploy distributed agent workflows
🛡️ Design for failure — retries, fallbacks, and human-in-the-loop checkpoints

Whether you are an AI engineer ready to ship your first agent or a developer curious about what autonomous AI systems really look like under the hood, this session delivers production-grade clarity — no hype, all code.