Observability pipelines were built for a world where humans wrote software and engineers manually standardized logs, traces, and metrics. That model is breaking down as AI begins generating services, libraries, and infrastructure faster than teams can maintain telemetry pipelines.  James Herbert (groundcover) explores a different model: observability layers that can reason about telemetry and adapt automatically as systems evolve. Rather than relying on engineers to constantly update pipelines, agent-driven workflows can continuously interpret and normalize telemetry as agentic and AI-driven infrastructure changes. This shift introduces new operational questions guardrails, validation, and how much autonomy modern systems should have.