Databricks prepares Genie for work that runs in the background
US data-and-AI company Databricks has announced a new always-on version of Genie for business tasks that require multiple steps and extended work. The agent is designed to track progress and request human decisions through a priority inbox and email. The new mode is headed for preview and beta programs, supported by company context and existing access controls.
Artificial Intelligence··Night
Genie is assigned continuing work
US data-and-AI company Databricks announced a new generation of Genie on 11 November for tasks that take multiple steps and extended work. The enterprise agent is designed to keep running in the background on its own computing resources. Employees supply a goal and boundaries, while Genie tracks progress, follows up on missing inputs and contacts a person when a decision is required.[1], [2]
A priority inbox is intended to collect matters needing human attention. Each new alert arrives as an unread email, with later updates to that alert retained in the same message thread. Databricks gives continuing tasks such as monitoring customer-use patterns, tracking revenue trends and maintaining a quarterly forecast as examples of the work that can be assigned to the agent.[1]
Business definitions provide context
Genie Ontology, Databricks’ system for representing company knowledge, supplies the definitions, metrics, people and processes that give a task business context. It draws on material including dashboards, queries and notebooks. The announced search capabilities extend across structured data and documents in business applications, including Google Drive, Jira and Confluence, using connections that can ingest information, index it in place or query it live.[1], [2]
The agent runs within an organization’s existing Databricks environment. Unity Catalog, the company’s data-governance system, applies access permissions, while Unity Gateway controls tool access, approval requirements and spending limits. Databricks says actions remain available for auditing. A planned Ontology Manager interface is intended to flag frequently requested concepts that need clearer definitions so teams can manage the context the agent uses.[1]
Internal use precedes wider testing of the new mode
Chief executive Ali Ghodsi demonstrated his own use of Genie to TechCrunch, including analysis of employee-survey weaknesses and a sales presentation that went poorly. The agent also reviewed earlier appearances to prepare him for a podcast interview. Those examples came from Ghodsi’s internal account rather than a staged product demonstration, showing the kinds of company information used in the tasks he described.[2]
Databricks distinguishes the established products from the newly announced operating mode. Genie and Genie Ontology are already generally available, while always-on Genie, Genie Search and Ontology Manager are headed for preview and beta programs. Other announced features include presentations and interactive reports grounded in changing company data, plus work with contracts, PDF documents and slides alongside governed tables.[1]
