Databricks has unveiled a new version of its Genie AI assistant that works on its own instead of waiting to be asked. Announced on 11 November 2026, the always-on Genie can be given an assignment or a goal and take ownership of the work, following up on missing information, delegating tasks to sub-agents and asking a person to step in only when a decision is needed.

The move is the centrepiece of a wider update to the Genie family. Databricks has also made Genie Ontology, the knowledge layer that gives its agents their understanding of a business, generally available, and introduced Genie Search, a governed way to search information across a company's whole data estate. Together, the company says, the changes turn Genie from a question-answering assistant into an AI coworker for long-running enterprise work.

From answering questions to owning the work

Until now, Genie has mainly responded when a user asked it something. The new version runs in the background on its own compute, keeping track of assignments over days rather than minutes. According to Databricks, it can monitor changes, request input when information is missing and manage several assignments at once, while users decide which actions need their approval before they happen.

The assistant can also hand parts of a job to sub-agents so that different strands of work run in parallel. It can also hand parts of a job to sub-agents so that different strands of work run in parallel. Databricks says every action stays within the governance controls a company already has in place, through its Unity Catalog and Unity Gateway systems.

Genie Ontology becomes generally available

At the heart of the update is Genie Ontology, which Databricks first introduced in June 2026 and has now made generally available. The ontology captures a company's context, including the definitions and metrics behind its data, wherever that data lives, and keeps it current as the business changes. It draws on tables, queries, dashboards, notebooks and connected workplace apps, so an agent knows, for example, which revenue figure the finance team trusts or what counts as an active customer.

Databricks argues that this shared understanding is what makes autonomy practical. Without it, an agent given a broad goal would have to guess at the meaning of the data it is working with. Analysts who cover the data platform market broadly agree that the context layer matters: Sanjeev Mohan, founder of analyst firm SanjMo, told TechTarget that agents fail less often from bad models than from missing context.

Genie Search and finished work in Genie One

The second major addition is Genie Search, a hybrid search tool that indexes the knowledge a company keeps outside Databricks, in documents, tickets and wikis held in tools such as Google Drive, Jira, Confluence, Workday and SharePoint. When a user asks a question, Genie draws on that index alongside its own data, and can also take live action through more than 100 connectors, such as checking the status of a Jira ticket or creating one.

Genie One, the home for this work, can now also turn an answer into finished work. It creates slide decks, dashboards and interactive pages directly in the conversation, with scheduled refreshes and shared skills. Genie is also reaching users where they already work, with native apps for web, iOS and Android, Slack and Microsoft Teams apps, a Chrome browser extension and desktop apps for macOS and Windows. Plugins bring Genie into ChatGPT, Claude, Microsoft Copilot Cowork and Grok Bot, using the same governed context and permissions.

Governance, pricing and a note of caution

Because the new Genie acts on its own, Databricks has put new emphasis on oversight. Unity Gateway, its governance layer, records a trace of what Genie did, who asked for it and why, and lets administrators set policies on which tools it may use and which actions need human approval. Databricks says Genie One is free for users through 31 January 2027, with more detailed cost controls to follow before then.

Analysts welcomed the shift while urging care. Mike Leone, an analyst at Moor Insights & Strategy, told TechTarget that handing Genie a goal instead of a question puts answers in front of people who would never open a dashboard, adding: "Taken together, I'd call [the new capabilities] significant because Genie stops waiting to be asked." Both Leone and Mohan said more transparency is still needed, including a clear way for the agent to say when it is unsure of an answer, and governance over what agents do in production, not only over what they can read.

Genie was first introduced in June 2024 as a conversational analytics assistant. For more detail, see the Databricks announcement and the TechTarget report.