Your AI Agent Is Confidently Wrong About Production — Willem Pienaar, Cleric

Why watch

Willem Pienaar explains why AI agents that excel at writing code fail at production debugging. The talk covers four key failure modes: mistaking symptoms for causes, missing baseline metrics, lossy sub-agent summaries, and anchoring on past incidents. He presents grounding techniques including parallel theories, dependency graphs, and outcome tracking to calibrate agent confidence. Ideal for AI engineers and DevOps practitioners building autonomous debugging systems.