An agent must answer questions about active service contracts stored as Dataverse rows and sometimes update a contract status field. Which design split is most appropriate?

Unstructured knowledge retrieval and structured CRUD against Dataverse are different integration modes. Updates require actions with proper security roles and connection references. The correct design uses Dataverse connectivity for transactional reads and status updates, while generative answers and knowledge patterns handle unstructured FAQs. The nightly CSV dump into knowledge files is stale for operational data and cannot perform secure updates. Storing status changes only in topic Message nodes does not persist business data. Putting production Dataverse connection strings in Adaptive Card visible text is a critical security failure. Exam tip: operational data uses Dataverse actions; documents use knowledge.