# Databricks launches autonomous Genie agent for long-running enterprise work

published: 2026-11-11T16:05:27.309Z
topic: frameworks
source: https://www.databricks.com/blog/new-genie-autonomous-ai-complex-enterprise-work
url: https://aiagentnews.io/stories/databricks-launches-autonomous-genie-agent-for-long-running-enterprise-work

Databricks has rebuilt Genie as an autonomous AI coworker that runs long-term enterprise tasks within guardrails set by the company's admins and users. The update adds Genie Ontology, Genie Search and Unity Gateway, which together handle business context, search across systems and governance of tool access and spending.

Genie now keeps working on ongoing jobs, tracking progress, following up on missing inputs and acting as new information arrives. Users can give it jobs through Genie One, email, Slack or Microsoft Teams, and it asks for input or approval when needed. Databricks cites examples such as monitoring insurance claims for fraud patterns and managing marketing campaigns with weekly budget proposals.

Genie Ontology holds definitions, metrics and relationships for a business, and agents build much of it from existing dashboards, queries and notebooks while people govern it. Genie Search combines an index of apps such as Google Drive, Jira, Confluence, Workday and SharePoint with live MCP search, and it enforces each source's access controls before returning results. Genie Agents can now also work with PDFs and slides through Unity Catalog Volumes without a separate vector index.

Unity Gateway lets admins control connector access, MCP tools and approval requirements, trace every action and set workspace and per-user budgets and spending caps. In a benchmark built with Turing, Databricks says Genie answered correctly in 91% of task runs, 1.6 times the rate of the stronger of two frontier agents it was compared with, while using less than half as many tokens. The company ran that comparison itself on the same tasks in the same Databricks workspace.

## Why it matters if you build agents

Developers building agents on Databricks can connect Genie to tools through MCP while admins set approval gates, tool limits and per-user spending caps. Teams that currently run a separate vector index for PDFs and slides may be able to drop it for Genie Agents. The accuracy and token figures come from a vendor-run benchmark on Databricks tasks, so they are worth testing against your own workloads.