Databricks Always-On Genie Coming to Preview Soon

Databricks on 11 November 2026 announced a second generation of Genie, described in a company press release as “an always-on, autonomous agent for enterprise work.” The same day’s Databricks blog, by Cynthya Peranandam and Elise Georis, sets out the architecture around that agent: Genie Ontology, Genie Search, and Unity Gateway as the control plane.

The availability line is more precise than the headline, and it is the one that should travel with the story. “Genie and Genie Ontology are generally available to all customers,” both official notices say. “Genie Search, Ontology Manager and the new always-on Genie are coming soon to preview and beta programs.” The always-on agent, the new search product and the ontology admin tool have been announced; they are not generally available, and Databricks has not said that preview access is open today.

What Databricks is promising, when those programmes open, is a Genie that no longer waits for a prompt. Employees, the press release says, “can give Genie a goal and trust it to keep work moving.” In Databricks’s account, Genie “takes ownership of the work, delegates to subagents when needed, tracks progress, follows up on missing inputs, and reaches out when a person needs to make a decision.” A new priority inbox “highlights what needs attention, and Genie emails users when they are away.” Those capabilities — subagents, the priority inbox and email, and the claim that the agent can run in the background on its own compute — are Databricks’s description of the forthcoming always-on mode, not features TSN can treat as shipping today.

What is already GA, and what is only coming soon

The distinction matters because Genie is not a new product name. Earlier this year TSN covered Genie One gaining code execution, the existing assistant that sits on warehouse data. The new blog says “thousands of customers use Genie in production,” and that Genie One remains the place teams can already work with configured agents. Always-on Genie is the second-generation mode Databricks is adding on top of that generally available assistant.

Genie Ontology, which is generally available, is described as “an agentic ontology built by agents and governed by people.” Databricks says agents build most of it automatically from dashboards, queries and notebooks, that it keeps learning with every query, and that teams can set authoritative definitions in Unity Catalog, with an algorithm called OntoRank giving those definitions precedence. Ontology Manager, by contrast, is “a new experience that lets teams manage and govern business context in one place,” and it is listed with the coming-soon preview and beta items, not with general availability.

Genie Search is also in that coming-soon group. Databricks says it “expands to 100+ enterprise data sources,” naming Google Drive, Jira and Confluence in the press release, and SharePoint and Workday in the blog, alongside lakehouse tables, dashboards and Databricks Apps. A related line says “Databricks now offers 100+ connectors” so teams can ingest data, index it in place, or federate queries through MCP. The “100+” figure is Databricks’s claim. The company also says Genie Search maps each source’s access controls in Unity Catalog “before returning any results, so nobody sees a document in Genie that they couldn’t open in the original application.”

Governance, Databricks says, already sits in Unity Catalog and Unity Gateway, including when the new agent mode arrives. Unity Gateway “lets admins control tool access, trace actions, and set spending limits.” Pricing is “pay for what they use, not per seat,” with those spending limits set “by team or by user through Unity Gateway.” Salesforce has also been adding spending caps and evaluation checks around enterprise agents; Databricks is making the same argument, that autonomy is only useful if an admin can bound it.

A separate Databricks claim, in a benchmark “built with Turing,” is that Genie, powered by Genie Ontology, “answered correctly in 91% of task runs, 1.6 times the rate of the stronger frontier agent, while using less than half as many tokens,” on tasks that required connecting knowledge across tables, dashboards and notebooks in the same Databricks workspace. Those figures are Databricks’s. TSN has no independent scores.

An always-on agent is still a preview promise

The announcement is best read as a product roadmap with a clear split, not as a general-availability launch of a background agent. Genie and Genie Ontology are already in customers’ hands. The always-on loop — an agent that keeps a job moving, writes to a priority inbox, and emails when a decision is needed — is still “coming soon to preview and beta programs,” as is Genie Search and Ontology Manager. Until those programmes open, the subagent, inbox and “100+ sources” claims remain Databricks’s account of what the next generation is for.

That is also why the comparison with other enterprise-table products is useful rather than decorative. NVIDIA’s open Kumo relational model is another bid to get predictions from company databases without first flattening every table into a text index. Databricks is arguing for a different shape: a governed ontology plus, when preview opens, a search layer over apps and an agent that will keep working after the chat window is closed. Ali Ghodsi, co-founder and chief executive, put the intended behaviour this way: “Every day, Genie keeps up with my inbox, preps me for meetings, and runs in the background, coming back only when it needs a decision.” That is the founder’s description of a mode that is not generally available yet.

The facts to watch are the opening of the preview and beta programmes, and whether Genie Search’s connector count and the Turing scores are later shown outside Databricks’s own write-up. Until then, the second-generation always-on Genie is an announced product, not a generally available one.