An honest look at the ways AI agents stumble — hallucination chains, confident mistakes, loops, and non-determinism — framed as normal workplace challenges with practical responses, not reasons to panic.