RESPONSIBLE AI, COMPLIANCE, AND RISK

CHECK YOUR UNDERSTANDING • What is a matched-pair bias test? ◦ Answer: Comparing results for two inputs that differ only in one detail, such as a name. • What is data minimization? ◦ Answer: Collecting and keeping only the personal data you truly need. • Why must audit records be tamper-evident? ◦ Answer: So any alteration after the fact can be detected. • What does the NIST AI Risk Management Framework offer? ◦ Answer: A voluntary guide for finding, measuring, and managing AI risks. • Why must accountability be assigned to people? ◦ Answer: A machine cannot be held responsible, so an organization must name who is. PART TEN: MULTI-AGENT SYSTEMS AND ORCHESTRATION In the first nine Parts of this book, you learned how to build and protect a single agent. An agent, as you know, is a software system built around an LLM that can take actions on its own toward a goal. Now we widen the lens. What happens when one agent is not enough, and you want several agents to work together? Part Ten answers that question in four chapters. Chapter 42 explains when and why to use more than one agent. Chapter 43 introduces the main patterns for organizing a team of agents. Chapter 44 explains the mechanics of passing work between agents. Chapter 45 covers how agents talk to each other, how to stop them from getting stuck, and how to oversee a whole fleet of them.