Overview
Databricks released the new Genie on November 11, 2026, turning it from a data Q&A tool into an always-on autonomous AI coworker. Two names need separating: Genie One is the entry point introduced earlier this year, while the new Genie is the subject here, reachable through Genie One, email, Slack, Microsoft Teams, the browser, desktop and iOS and Android.
The behaviour change is that it takes responsibility. Give it a job and access to the tools it needs and it keeps following the work, tracking progress, asking when it needs input and acting as new information arrives, running over days and weeks. The examples Databricks gives make it concrete: monitoring incoming insurance claims for emerging fraud patterns and loss-ratio drift, assembling the case file and routing it to underwriters when a segment moves outside guidelines; or managing campaigns around the clock, investigating underperformance and proposing weekly budget changes for approval.
Proactivity is designed in rather than incidental. It notices work you repeat, such as a weekly update, and offers to take it on, and it can flag issues needing attention before you set up a job at all. Jobs run on a schedule or from triggers, such as a metric dipping or more than 10 customers asking for the same feature. It runs many jobs at once, and you supervise, steer and stop them in one conversation with a panel showing active jobs and outputs.
Three pieces hold this up. Genie Ontology is the business ontology, telling it what counts as an active customer or which revenue metric is authoritative; agents build most of it from existing dashboards, queries, notebooks and semantic models such as Power BI, people establish authoritative definitions in Unity Catalog, and the OntoRank algorithm gives those precedence so every agent works from the same governed understanding. Genie Search combines an index of Google Drive, Jira, Confluence, Workday and SharePoint content with live retrieval over MCP, enforcing each source's permissions before returning results, so nobody sees a document in Genie they could not open in the original app. Unity Gateway is the control plane for connector access, MCP tools and approval requirements, with workspace and per-user budgets, spending caps, alerts and traceable audit of every action.
Unstructured data is now covered. Genie Agents work across PDFs and slides as well as structured tables through content search in Unity Catalog Volumes, with no separate vector index and page-level citations for checking evidence.
Connectors reach Slack, Zoom, Microsoft 365, Jira, GitHub and calendars, with custom MCP for systems like ServiceNow and PagerDuty. One concrete pattern: investigate an incident using GitHub and Jira changes, keep the ticket fresh and update the team until the issue resolves.
Distribution runs on two levels. Genie plugins carry business context into ChatGPT, Claude, Copilot and Grok Bot, and Genie MCP connects other AI tools directly. OpenSharing lets teams hand governed agents preloaded with data, instructions and knowledge to partners and customers, while Genie Clean Rooms let recipients ask questions and get answers without exposing underlying data or proprietary logic, with limits on conversations and outputs.
Read the benchmark carefully. Databricks built an enterprise data agent benchmark with Turing, comparing Genie against two leading frontier agents on the same tasks in the same live Databricks workspace, where tasks required connecting knowledge spread across tables, dashboards and notebooks rather than simple data access. Genie answered correctly in 91% of task runs, 1.6 times the stronger frontier agent, using less than half the tokens. It is a vendor-built benchmark with the vendor winning, so treat the direction as credible and the absolute number with room to spare.
The customer list is mild corroboration: Zillow, Unilever, adidas, Atlassian and AstraZeneca run Genie in production, and Databricks says thousands of customers use Genie One.
Key Features
- Holds long-running jobs: Give it a job and tool access and it keeps following the work, tracking progress, asking for missing inputs and acting as news arrives, on a schedule or from triggers.
- Proactive pickup: Notices repeated work and offers to take it on, and can flag issues needing attention before a job is even configured.
- Many jobs, one supervisor view: Runs multiple jobs at once and lets you supervise, steer and stop them in one conversation, with a panel for active jobs and outputs.
- Genie Ontology: Agents build the business ontology from dashboards, queries, notebooks and semantic models such as Power BI, people set authoritative definitions in Unity Catalog, and OntoRank orders by authority for every agent to share.
- Genie Search: Combines an index with live MCP retrieval across Google Drive, Jira, Confluence, Workday and SharePoint, enforcing each source's permissions before results return.
- Unity Gateway: Governs connector access, MCP tools and approval requirements, with workspace and per-user budgets, spending caps, alerts and traceable audit.
- Unstructured data: Genie Agents handle PDFs and slides through content search in Unity Catalog Volumes, needing no separate vector index and returning page-level citations.
- Artifacts and reusable skills: Describe a deck, dashboard or interactive HTML page and Genie builds it in the conversation, keeps it current on a schedule, and skills live in Unity Catalog running with the user's permissions.
Use Cases
- Monitoring insurance claims for fraud patterns and loss-ratio drift
- Around-the-clock campaign management with weekly budget proposals
- Business attribution across tables, dashboards and notebooks
- Productising departmental expert agents
- Investigating incidents from GitHub and Jira changes while keeping tickets current
- Carrying governed business context into ChatGPT, Claude and Copilot
- Sharing governed agents with partners through OpenSharing and Clean Rooms
Pros
- Holds long-running jobs, asks for missing inputs and acts under guardrails or requests approval, which suits work spanning days and weeks
- Genie Ontology is built by agents from existing assets while people set authoritative definitions in Unity Catalog, ordered by OntoRank
- Genie Search mixes an index with live MCP retrieval and enforces each source's permissions before returning results
- PDFs and slides are searchable through Unity Catalog Volumes without a separate vector index, with page-level citations
- Unity Gateway governs tool access, spending caps and audit in one place with workspace and per-user budgets
- Plugins carry the same context into ChatGPT, Claude, Copilot and Grok Bot without rebuilding it
Pricing
A paid Databricks platform product billed by workspace, with specific tiers confirmed through sales. Unity Gateway lets administrators set workspace and per-user budgets, spending caps and alerts, with traceable audit of every action.
Summary
The hard part of an enterprise agent is not whether the model can analyse. It is whether the agent knows how your business defines things, respects your permissions and has someone watching the spend. The new Genie ships all three: an ontology for business understanding, search that pulls scattered systems together and enforces permissions before returning anything, and a gateway for tool access and cost. It has moved from answering questions about data to holding down ongoing work, following up on missing inputs and running for days. The 91% task pass rate comes from a benchmark Databricks built with Turing, so read it as a strong signal on a home-field test.
Version History
- New Genie released as an autonomous AI coworker (2026-11-11): Databricks released the new Genie, moving from data Q&A to an always-on autonomous AI coworker that takes jobs through Genie One, email, Slack, Teams and mobile, follows long-running work, asks for missing inputs and acts under guardrails or requests approval, with concurrent jobs under one supervisor view. Launched alongside Genie Ontology (built by agents, governed in Unity Catalog, ordered by OntoRank), Genie Search (index plus MCP retrieval with permissions enforced before results return) and Unity Gateway (tool access, spending caps, audit). Genie Agents added unstructured analysis of PDFs and slides plus MCP connectors. On an enterprise data agent benchmark built with Turing, Genie passed 91% of task runs, 1.6 times the stronger frontier agent, using less than half the tokens.