Agentic analytics typically relies on an architecture consisting of data in a predictable format that agents can reason over, a semantic layer describing business context, and AI agents that query this data foundation. 

This doesn't necessarily mean structured data. Agents can work with unstructured data too, for example in log analysis or support operations. What matters is that the data is in a consistent, well-organized format. Behavioral data is captured and processed at the point of collection, stored in a data platform such as a warehouse or lakehouse, and described through semantic metadata. AI agents then interact with this data foundation to generate insights and recommendations.