Overview
- Databricks announced that Genie can now run long-running, multi-step jobs unsupervised and follow up for missing inputs while letting admins require approvals for actions.
- Genie Ontology and Genie Search were declared generally available to supply structured context from Unity Catalog pages, metric views and linked documents so the agent cites governed definitions instead of guessing.
- The company also released control tools through Unity Gateway that trace questions and actions, enforce tool permissions, require approvals, and set per-user and per-workspace spending caps.
- Databricks published a vendor benchmark reporting 91% task accuracy for Genie in a mature Turing-powered deployment but cautioned the result reflects a tuned, established workspace and may not match new or uncurated environments.
- Several features including document-reading agent capabilities, the visual Ontology Manager and some integrations remain in beta or preview, so teams are advised to start with read-only jobs, write governed definitions first, and enable tracing and approval rules before letting Genie act autonomously.