05 execution rules and failure management

Execution Rules and Failure Management explains how managers can ensure that AI agents operate reliably, safely, and accountably by establishing clear execution rules, operational boundaries, and failure management procedures. It emphasizes defining required inputs, expected outputs, workflow triggers, validation checks, and completion conditions before allowing agents to perform tasks. Managers must also establish filtering rules, task priorities, processing limits, and permissions to prevent unauthorized actions or incorrect decisions. The chapter introduces three important failure management techniques: retries, which address temporary technical problems; graceful degradation, which allows reduced but useful service when certain functions fail; and circuit breakers, which suspend operations when repeated failures occur. It further highlights the importance of human escalation when information is missing, policies conflict, or decisions require managerial judgment. Using HR onboarding examples, the chapter demonstrates how AI agents should detect errors, prevent duplicate actions, maintain accurate records, and request human intervention when necessary. Finally, it emphasizes maintaining decision trails and audit records that document inputs, validation results, actions, approvals, and outcomes while protecting sensitive information. The central message is that reliable AI systems require not only effective task execution but also clearly defined procedures for identifying failures, controlling risks, recovering safely, and maintaining human accountability.