This is a really important distinction that most teams overlook. I've been building automation agents for clients and the biggest pain point is always stale data — an agent makes a decision based on cached state, and by the time it acts, the context has shifted. Materialized views are a solid pattern here. Curious if you've seen teams pairing streaming databases with tool-calling LLM agents directly, or if there's usually a middleware layer handling the freshness guarantees?