The problem is the source

I have been reading Observability Engineering, 2nd edition. A point that stays with me is that an AI-agent looking at telemetry can query anything and still “understand” nothing (if an agent can even be said to understand). The knowledge needed to interpret the signals usually lives in human heads. What the service is for, what normal looks like, why that threshold exists, what we tried last time, what tends to fail and what looks like a failure but is actually normal. Under the heading “The Agentic Incursion Has Just Begun” (ominous?) I found this section: Agents running in production can only be as good as the data they have access to. … Most of all, they need us to think critically about the intuitive leaps we make and convert them into breadcrumbs of context, cardinality, and other clues. - Observability Engineering, 2nd edition, Majors et al., 2026 That is written about production debugging. I think converting our intuitive leaps into committed breadcrumbs has a larger impact…