Moving AI agents from pilot demos to enterprise production requires a fundamental shift in operational monitoring. Unlike traditional deterministic applications, AI agents are non-linear, stateful, and autonomous—frequently calling external tools, branching on intermediate reasoning, and coordinating across multi-agent workflows. Standard log metrics simply fall short.
In this session, explore how Azure AI Foundry bridges this gap by unifying agent execution with enterprise-grade observability via OpenTelemetry and Azure Monitor Application Insights. We will break down the spectrum of tracing approaches—from no-code out-of-the-box visibility with Microsoft Agent Framework, to low-code tracing with Semantic Kernel, LangChain, and LangGraph, to pro-code custom instrumentation with OpenAI Agent SDK. You’ll leave with actionable architecture patterns to achieve end-to-end visibility into token usage, tool invocations, decision trees, and multi-agent interactions in production.