Loading page…
The update combines business context, connected applications and approval controls. Databricks also claims an accuracy advantage in a benchmark built with Turing.
Listen to this story
Databricks is positioning Genie as a supervised operator for work that continues across enterprise tools, not just a chat interface: employees can schedule jobs or trigger them from changing conditions, then monitor, steer or stop them. The approach depends on business context assembled from company data and documents, with administrators governing tool access and employees choosing which actions need approval. Databricks reports strong benchmark results, but those figures come from its own comparison of tasks in a Databricks workspace.
Genie Ontology and Genie Search became generally available November 9; staged rollouts may take a week or more to reach individual accounts.
Databricks reports 91% correct task runs and a 1.6-times higher correct-answer rate than the stronger comparison agent, using less than half as many tokens.
MCP connections let Genie read and act in Slack, Microsoft 365, Jira and GitHub.
Databricks introduced a new version of Genie on November 11, 2026, designed to keep working after an employee hands over an assignment. The company says its AI coworker can track changes, chase missing inputs and act across business tools on long-running enterprise tasks, while people set limits and decide which actions need approval.
Two foundations arrived just before the announcement. Databricks’ release notes list Genie Search and Genie Ontology as generally available on November 9. Releases are staged, so individual accounts may receive updates a week or more after the initial release date.
Genie’s jobs can run on a schedule or start when something changes, such as a falling metric. Employees can supervise, steer and stop several jobs in one conversation, with a control panel showing active work and outputs.
Databricks says Genie can notice repeated work, such as a weekly update, and offer to take it on. It can also flag issues before someone creates a job.
The company illustrates the intended work with marketing campaigns: Genie could investigate underperformance and propose weekly budget changes for approval. Connections using Model Context Protocol, or MCP, let it read information and take actions in tools including Slack, Microsoft 365, Jira and GitHub.
Genie Ontology connects data with business definitions and relationships. Databricks says agents build much of it from dashboards, queries and notebooks. Teams establish authoritative definitions in Unity Catalog, its governance system; an algorithm called OntoRank gives those definitions precedence.
Genie Search retrieves information from sources including Google Drive, Jira, Confluence and Workday. Databricks says it keeps document permissions current alongside content, restricting results to documents users can already access. Specialized Genie Agents also gain access to PDFs and slides alongside tables, with page-level citations.
Databricks’ benchmark, built with Turing, compared Genie powered by Ontology with two frontier agents in the same Databricks workspace. Tasks required connecting knowledge across tables, dashboards and notebooks, rather than simply finding data.
Databricks reports 1.6 times the correct-answer rate of the stronger comparison agent, using less than half as many tokens.
For administrators, Unity Gateway and Unity Catalog govern tool access and permissions, expose actions and usage, and support spending limits, Databricks says. Employees choose which steps require sign-off.
Loading discussion...
Join the conversation
Explain when that initiative would feel useful rather than distracting.
Be the first to share a perspective or an experience.
Reader comments
Newest comments first. Replies stay oldest first.