As AI is integrated into critical applications and autonomous agents, the need for observability becomes key to ensuring quality, reliability, and compliance. In this talk, we will explore how Azure AI Foundry enables instrumentation of AI models, apps, and agents to capture traces, metrics, and evaluations that provide comprehensive visibility into system behavior.

We will introduce Azure AI Foundry's unified approach to observability, covering everything from instrumentation with the SDK to analysis and visualization in the portal. We will look at how to monitor interactions, detect deviations, assess response quality (from humans or models), and close the loop with dashboards and alerts.

The session will include a hands-on demonstration where we will instrument an application based on generative models and agents, demonstrating how to achieve end-to-end traceability to facilitate diagnostics, continuous improvement, and responsible governance of AI solutions.