How to Instrument, Govern, and Debug Agents Before They Go Rogue

Agents don’t just need prompts, they need supervision. As teams move from experiments to production, they often discover the hard way that generative systems are unpredictable, opaque, and prone to silent failure. Observability isn’t just a nice-to-have, it’s the only way to keep your agents aligned, your users safe, and your systems sane. In this talk, we’ll dive deep into the emerging discipline of LLM observability: what to track, how to track it, and why it matters. We'll compare purpose-built tools like LangFuse, LangSmith, and AgentOps with general observability frameworks like OpenTelemetry, and show how new GenAI semantic conventions make structured tracing possible at scale.