Databricks lets Genie Agents read documents and act: unstructured analysis hits public preview as MCP connectors go GA

Genie Agents can now work over governed tables and large document sets at once, and act inside enterprise tools such as Slack, Google Drive and Microsoft 365 by creating tickets, updating records or posting messages. The product had been positioned as the fastest way to turn business questions into SQL insights, relying on structured business data alone. Databricks describes an agent pulling margin and selling-price trends from governed tables, then searching attached analyst reports and earnings transcripts to return data trends, likely causes and page-level citations for each claim.

Document handling is automatic: the agent parses, indexes and retrieves context, so builders no longer assemble a retrieval-augmented generation pipeline themselves. Large-scale document analysis previously required custom advanced parsing, a separate AI Search endpoint, an embedding pipeline and summaries passed to a supervisor agent framework with permissions, making build and maintenance costly. A custom AI document parsing function reads files as a person would, recognizing tables, figures, headers and footers while preserving context; parsing and indexing run at scale so one agent can work through thousands of files and cite the page behind each answer, running in Agent mode under Unity Catalog permissions.

The report covers only the product capabilities and customer use cases Databricks announced: unstructured data analysis is in public preview, native MCP connectors are generally available, and OpenSharing and Genie Clean Rooms are in beta; no quantitative comparison of accuracy, latency or cost is given.

The announcement argues value through vendor-described customer use cases, without disclosing error rates for document parsing on complex layouts or quantified privacy-preserving effects of cross-organization sharing; Genie Clean Rooms and OpenSharing remain in beta and may change.