Monitoring traditional microservices is a solved problem. We know how to track CPU, memory, and latency. But how do you monitor the "reasoning" and execution loops of an AI agent? If you can't measure the token cost, logic paths, and hallucination rates of your GenAI workflows, you are flying blind in production.
This session explores how to adapt the open-source monitoring tools you already use—specifically Prometheus and OpenTelemetry—to track the health and infrastructure costs of agentic workflows. We will treat LLM calls just like any other distributed microservice dependency. I will demonstrate how to instrument an AI pipeline, map token usage directly to infrastructure spend, and build Grafana dashboards that expose both system health and model performance in real-time. Stop guessing what your agents are doing and start monitoring them like proper engineering assets.