Databricks Unveils Always-On Genie — But the Always-On Part Is Preview Only

Databricks has unveiled the second generation of Genie, its AI agent for enterprise work — and the headline feature, the "always-on" mode that lets Genie work in the background on its own, is not something customers can use yet. The announcement, made in San Francisco on 11 November 2026, splits the release in two: the core Genie product is generally available, but the always-on mode is coming soon to preview and beta programmes, with no dates given. Read the announcement carefully and the picture is clear — the future Databricks is selling is real, but the most interesting part of it is still behind a waiting list.

The GA vs Preview Split — Read This First

Databricks' own words draw the line. "Genie and Genie Ontology are generally available to all customers. Genie Search, Ontology Manager and the new always-on Genie are coming soon to preview and beta programs." Generally available, or GA, means any customer can turn it on today. Preview and beta mean a limited group gets access first, and everyone else waits. The always-on mode — the thing the announcement is named after — is in the second group.

Capability	Status (11 Nov 2026)
Genie	Generally available to all customers
Genie Ontology	Generally available to all customers
Genie Search	Coming soon to preview and beta
Ontology Manager	Coming soon to preview and beta
Always-on Genie	Coming soon to preview and beta

What "Always-On" Actually Means

So what is the always-on mode, and why does it matter? A "second-generation" Genie is, in Databricks' words, an "always-on, autonomous agent for enterprise work". The key word is autonomous. Today's data assistants wait for you to ask a question. Genie in always-on mode does not wait. It runs in the background on its own compute, delegates work to sub-agents, tracks progress, follows up on its own, and reaches out to people through a new priority inbox and email when it needs a decision or has finished something.

Think of the difference between an assistant you summon and a colleague who works while you are in another meeting. The first is useful when you remember to ask. The second is useful because it notices what needs doing. That is the shift Databricks is selling: from a chatbot that answers data questions to an agent that owns data work end to end — monitoring, preparing, checking and delivering, with humans stepping in only where judgement is needed.

AI-generated illustration by INORYVA.

What Is Generally Available Today

The parts of Genie that customers can use right now are substantial on their own. Genie itself is Databricks' conversational AI for data: ask a question in plain language, and it queries your company's data and returns an answer, a chart or a report. Genie Ontology is the layer underneath it — the governed business context that tells Genie what your company's terms mean, so that "revenue" in a question means the same thing as "revenue" in the finance team's reports.

Genie Search is the other big expansion, and it is in the preview group. It extends Genie to more than 100 enterprise data sources, covering the lakehouse alongside Google Drive, Jira, Confluence and other systems where companies actually keep information. A question that needs a Jira ticket, a Drive document and a warehouse table can, in principle, be answered in one place. That is the kind of capability that sounds small in a press release and is large in practice — if it works across all those sources with the same accuracy.

Governance: The Part That Has to Work

Databricks is putting its governance story front and centre, and for good reason. Genie runs inside the customer's existing Databricks environment, and every action it takes sits within Unity Catalog permissions and Unity Gateway approval rules and budgets, with a full audit trail. Unity Catalog is Databricks' system for controlling who can see and use each piece of data; Unity Gateway is the control layer for AI usage, including spending limits. In plain terms: Genie can only touch data its user is allowed to touch, can only spend what an admin has approved, and everything it does is recorded.

That design is what makes always-on mode plausible at all. An autonomous agent without those limits would be a liability — a system that could email the wrong person, query restricted data, or run up a compute bill while nobody is watching. The audit trail matters too: when an agent does something in the background, someone has to be able to reconstruct what it did, what data it used and why. Databricks is betting that enterprises will accept autonomous agents only if the controls are at least as strong as the ones they already use for their human analysts.

Connections and Pricing

On pricing, Databricks' FAQ is clear about the model and silent about the number that matters most. "Genie is part of the Databricks Data + AI Platform. Customers pay for what they use, not per seat, and admins can set spending limits by team or by user through Unity Gateway." Usage-based pricing, no per-seat charge, admin-controlled limits. What Databricks does not state is any specific price for always-on mode — so this article quotes none.

One secondary pricing detail is worth noting with a flag, because it is circulating but does not come from Databricks' own announcement: Genie One and Genie Agents LLM usage is said to be free until 31 January 2027, with each user then getting 150 DBUs a month free from 1 February 2027, and $0.070 per DBU beyond that, plus SQL warehouse compute. That has not been confirmed in Databricks' own statement for this release, so treat it as unverified until Databricks says it itself. If you are budgeting for Genie, ask Databricks directly rather than relying on a second-hand price.

AI-generated illustration by INORYVA.

Why This Launch Matters Anyway

The release also shows how the AI agent race is splitting. One camp is building agents for consumers and developers — coding assistants, chatbots, decision APIs. Databricks is building for the enterprise data team, where the hard problems are not intelligence but governance, accuracy and cost control. That is a less glamorous contest and a more valuable one: the company that can let an agent loose on a business's data without breaking its rules will win the budgets that matter.

AI-generated illustration by INORYVA.

What to Watch Next

The honest test of this announcement is the preview. When always-on Genie opens to preview and beta customers, watch for three things. First, who gets in and how quickly — a narrow, slow preview suggests the capability is further from ready than the announcement implies. Second, what the first users report: does Genie actually complete multi-step work in the background, or does it need constant supervision? Third, the governance in practice: do the Unity Catalog and Unity Gateway controls hold up when an agent is acting on its own, and does the audit trail satisfy the compliance teams who will have to approve it?

Key Takeaways

• Databricks announced second-generation Genie on 11 Nov 2026: an "always-on, autonomous agent for enterprise work" that runs in the background, delegates to sub-agents and follows up via a priority inbox and email.
• The always-on mode is NOT generally available — Databricks says it is "coming soon to preview and beta programs", with no dates given.
• Generally available today: Genie and Genie Ontology. Also preview/beta only: Genie Search and Ontology Manager.
• Genie Search expands to 100+ enterprise data sources, including the lakehouse plus Google Drive, Jira and Confluence.
• Governance runs through Unity Catalog permissions and Unity Gateway approval rules and budgets, fully auditable, inside the customer's Databricks environment.
• Pricing is usage-based, not per seat; no specific price for always-on mode is stated. A secondary free-until-31-Jan-2027 pricing claim is unverified and not from Databricks' own release.
• Genie connects via MCP to ChatGPT, Claude, Copilot Cowork and Grok Bot, and Genie Agents now work across PDFs, contracts and slides.

Conclusion

Always-on Genie is the most interesting idea Databricks has announced for its AI assistant — an agent that works while you do something else, instead of waiting to be asked. But the idea is ahead of the product: the always-on mode is preview-only, with no date, and the governance that makes it safe is still to be tested by real customers. Databricks deserves credit for drawing the GA-versus-preview line clearly in its own announcement. The rest of us should read it just as clearly, and judge Genie by what the preview proves, not by what the headline promises.

Sources

Databricks — Databricks Launches New Always-On Genie, Autonomous AI for Complex Enterprise Work