From chaos to clarity: Patterns for generative AI observability

AI is being adopted at a rapid pace, and MCP and agents are becoming a new standard in software systems and platforms. Does the team know what is happening to these workloads? AI workloads present unique challenges such as token usage, event tracing, and model performance. As nondeterministic systems, LLM outputs can be unpredictable, so tracking input data and event triggers can help explain how a prompt led to a decision.

Observability with generative AI should not be an afterthought. In this talk, attendees will learn best practices in AI workload observability based on usage patterns from teams that have implemented prompt logging, tracing, and similar techniques.