ENDING A RUN

CHECK YOUR UNDERSTANDING • Why not trust the agent's own claim of being done? ◦ Answer: Models can sound confident while being wrong, so completion needs an independent check. • Name the four parts of a structured failure report. ◦ Answer: Status, reason, partial progress, and recommended next step. • Why pass structured output between agents? ◦ Answer: The next agent needs only key findings, and a small clean context works better. • When should cleanup run? ◦ Answer: Always, including when the run fails. • What is partial success? ◦ Answer: Some goals were met and others were not. PART SEVEN: THE OBSERVABILITY LAYER (SEEING INSIDE THE AGENT) A doctor cannot treat what she cannot see. That is why hospitals have thermometers, heart monitors, and X-ray machines. An agent run is a hidden process, with reasoning, tool calls, and decisions happening inside software. The observability layer is the set of instruments that lets engineers see inside. In this Part you will learn why observability matters (Chapter 27), how tracing and spans record each step (Chapter 28), how logs, metrics, and dashboards watch the whole system (Chapter 29), and how replay helps you debug (Chapter 30).