Do we really need LLM observability?

- Operationally Intelligent