Imagine a procurement agent doing exactly what it was supposed to do. A supplier flags a delay. The agent reads the email, finds the affected PO, scans the network for alternate inventory, and reroutes the order. Twelve seconds, end to end. In a demo, the room nods. Someone asks about hallucinations. The vendor says the right things about guardrails. Everyone walks away reassured. The interesting question is a different one. Not whether the agent could be wrong β but what happens on the day it's completely, devastatingly right. The failure mode nobody is demoing: A financial agent told to minimise cost on a category executes a renegotiation perfectly. Margin is squeezed. Terms are tightened. The supplier, who was already thin, collapses six months later. The agent didn't malfunction. It succeeded. The metric was the bug. This isn't a hallucination.β¦