WHY OBSERVABILITY MATTERS

EXAMPLES Example 1: The hidden middle A user asked a question and received a wrong answer. The input-output log shows only those two items. The trace shows fourteen steps between them, including a search that returned an unrelated page, which is where the error began. Example 2: A stale cache An agent repeatedly cites an outdated price. Observability data shows the tool returned a cached result from last week. The team refreshes the cache more often, instead of wasting time retraining or rewriting prompts. Example 3: A pattern in failures A dashboard shows that thirty percent of failed runs share one tool-call error. Engineers fix that tool first and gain the largest improvement. Example 4: Two complaints, one cause Two users reported "it just didn't work" on different days. Full tracing showed both runs failed at the same database lookup. The shared root cause was found in under five minutes. PRACTICE LAB CONNECTION Add tracing and logging to the Resume Review Agent lab in Chapter 68. Then test the prompts below against its run logs.