WHY OBSERVABILITY MATTERS

ARTIFICIAL INTELLIGENCE ASSISTANT PROMPTS [START OF PROMPT] We only have input/output logs for this failure, no tracing: [PASTE USER REQUEST AND FINAL ANSWER HERE]. Identify what observability data (spans, intermediate reasoning, tool calls) is missing that would be needed to actually diagnose whether the answer is right for the wrong reason, and specify exactly what logging/tracing instrumentation to add to our harness so this type of failure is diagnosable next time. Output format: a diagnosis-gap note plus a specific instrumentation change list. [END OF PROMPT] [START OF PROMPT] Here is a vague user complaint with no detail: [PASTE COMPLAINT HERE]. Given our available observability data: [PASTE AVAILABLE TRACES/LOGS/METRICS HERE], produce a diagnostic plan: what to check first, second, third, and the most likely root cause given the evidence. Output format: a step-by-step diagnostic plan followed by a most-likely-root-cause hypothesis with supporting evidence. [END OF PROMPT] CHAPTER SUMMARY Observability is understanding a system's inside from the signals it sends out. Input-output logging hides the many middle steps of an agent run. Observability data reveals the true root cause, which decides the right fix. Patterns in failures show where a single fix gives the greatest benefit.