In November 2022, a passenger named Jake Moffatt turned to Air Canada’s support chatbot to ask about bereavement fares for a last-minute flight to a funeral. The AI, helpful, polite, and confident, explained that he could book a full-price ticket now and claim a refund within 90 days. Moffatt followed these instructions. But when he later attempted to claim that refund, the airline denied it, stating that the policy described by the chatbot simply did not exist.

When the dispute reached a civil resolution tribunal in 2024, Air Canada argued that the chatbot was a “distinct legal entity” and that the airline could not be held responsible for the machine’s autonomous statements. The tribunal rejected this defense and ordered the airline to pay (Moffatt v. Air Canada, 2024 BCCRT 149).

This case shattered the illusion that AI is magic. It demonstrated that in a professional context, a large language model (LLM) that “hallucinates” is not a quirky conversationalist; it is a liability. For builders and professionals, this is the reality of the tool you are wielding. You are not chatting with a smart assistant. You are operating a probabilistic engine that is eager to please, statistically likely to sound correct, and entirely capable of confidently lying to your customers.

This book is the manual for preventing such failures. It moves beyond the casual “chat” mindset to establish a framework for engineering reliable, production-grade instructions.