When your "new teammate" hallucinates: Observability as the safety net for AI code

As AI copilots and coding agents generate more of the code teams ship, they can deliver software faster than they can understand it, leading to fragile behavior, missed edge cases, and failures that are hard to explain.

This talk's guiding principle is simple: the faster teams generate code, the more they need telemetry. It connects observability fundamentals, including signals, context propagation, and debugging workflows, to the realities of AI-generated code, using OpenTelemetry as a vendor-neutral foundation for traces, metrics, and logs that reveal runtime truth.

In this session, attendees will learn how observability tools can help them stay sane and in control while adding new features and functionality with AI-generated changes. They will gain a clear understanding of how to use "instrumentation by prompt," autoinstrumentation, and manual instrumentation to make code reliable and understandable.