Databricks launches always-on Genie AI agent for enterprise work

Databricks has launched the second generation of Genie, an always-on autonomous AI agent designed to run continuously inside a company's existing data environment and handle long-running work without repeated prompts. The release, announced 12 November 2026, marks the company's most aggressive push yet to embed AI agents into corporate data and software systems — and it comes with governance controls baked in, because that is exactly where large organisations have been hesitating.

The core shift is simple to state: Genie no longer waits to be asked. A worker assigns a goal, sets boundaries, and the system tracks progress, follows up on missing inputs, and returns only when a human needs to make a decision. That is a meaningful departure from the question-and-answer pattern that dominates most enterprise AI tools today.

What does always-on actually mean?

Under the new model, Genie runs in the background and takes on operational work rather than one-off queries. Databricks lists concrete use cases: monitoring quarterly forecasts, tracking revenue trends, analysing employee survey responses, and identifying changes in customer usage patterns.

The mechanics matter here. Genie can delegate work to sub-agents, present outstanding issues in a priority inbox, and send email alerts when it needs input from a person. So the agent is not just answering questions — it is managing a queue of ongoing responsibilities and escalating selectively.

That is the tension Databricks is targeting. Many companies are experimenting with AI tools that can answer questions or complete discrete tasks, but they remain wary of giving those systems access to financial planning data, customer records, and internal documents. Databricks' answer is to keep the agent inside the customer's existing environment, subject to the security, permissions, and approval controls already set through Unity Catalog and Unity Gateway. Every action is auditable.

The takeaway: Databricks is betting that enterprises will trust autonomous agents only when the agents inherit the company's existing permission structure — so that is exactly what Genie does.

How does Genie Search work?

The launch also includes Genie Search, a governed search layer that spans both structured and unstructured information, built for both staff and software agents. It extends the product beyond the Databricks lakehouse into business applications including Google Drive, Jira, and Confluence.

Teams have three options for getting content into the layer:

• Index content where it already sits.
• Ingest it through Lakeflow Connect.
• Query applications live through MCP.

Databricks now offers more than 100 connectors to support those approaches. The practical effect, according to the company, is that users and agents can search once across multiple sources instead of searching each application separately — which reduces tool calls and token usage. For companies running large agent fleets, token economics is not a side detail; it is a line item.

The takeaway: a single governed search layer across scattered business apps is meant to cut both friction and cost, and Databricks says it does both.

What is Genie Ontology?

The search function is tied to Genie Ontology, Databricks' framework for capturing business definitions, metrics, reporting lines, and processes. The company describes it as a way to maintain an up-to-date understanding of how a company defines its data and how its people and processes fit together.

Two additions ship with this release:

• Support for business logic from external business intelligence tools.
• A new organisational structure connector.

Databricks also introduced Ontology Manager, a new interface for teams to manage and govern business context. The tool can recommend areas for refinement by flagging concepts that are frequently queried — a maintenance loop that keeps the ontology from going stale as the business changes.

The takeaway: shared definitions are the plumbing underneath reliable AI answers, and Ontology Manager gives teams a dedicated tool to keep that plumbing clean.

Where can people use it?

Distribution is the other half of the story. Users will be able to access Genie through a web browser, a mobile app, email, and a range of collaboration and AI products, including ChatGPT, Claude, Microsoft Copilot Cowork, and Grok Bot. Meeting users inside the tools they already work in — rather than forcing a new destination — has become the standard playbook for enterprise AI, and Databricks is following it.

The release also broadens what Genie agents can work with. Agents can now operate across PDFs, contracts, and slide decks stored in Unity Catalog volumes alongside governed tables, without the need for a separate vector index. That removes a step that previously added setup cost and complexity.

Databricks is also adding dynamic slides, dashboards, and interactive reports that stay tied to governed data and update as that data changes. The point is continuity: a report is no longer a frozen artefact but a live view of numbers the company has agreed to trust.

The takeaway: Genie is spreading outward — into more tools, more file types, and more live surfaces — while staying anchored to governed data.

What do the customers say?

Ali Ghodsi, Co-founder and Chief Executive Officer of Databricks, said the product has already changed how he works. "Genie has made me a 10x more productive CEO, and now it doesn't wait for me to ask," Ghodsi said. "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. It can do that because it knows how our company works: what we mean by revenue, which numbers are official, and who can see what. That's the power of horizontal context, and Genie delivers it."

Partners framed the same idea in operational terms. Jaime DeLanghe, Chief Product Officer of Slack, said governed context is what makes agents useful inside everyday work. "Teams make their best decisions when trusted context shows up in the flow of work. With Genie Agents now connected to Slack, sales, finance, and operations teams can go from question to action without leaving the conversation," DeLanghe said. "We're thrilled to work alongside Databricks to bring governed, reliable insights to where the work happens."

Headspace pointed at definitions as the fix for inconsistent AI output. "AI is only as good as the definitions behind it, so we built our Genie Agents on agreed, governed definitions, and now every answer matches," said Marlissa Wong, Director of Data Engineering at Headspace. "We expanded self-service analytics by putting plain-language access to our data in the hands of teams across the business, not just analysts, so people can act on numbers they trust."

HelloFresh brought a hard number. "With Genie constantly analysing performance signals, we've improved programmatic ad ROI and reduced the time to identify ad winners on Meta by 75%," said Alan Khalil, Associate Director of Data Science and Engineering at HelloFresh. "Unity Gateway gives us the flexibility to use the best model for each job without managing separate integrations."

A 75% cut in the time to identify winning ads on Meta is the kind of figure that moves procurement conversations, and Databricks clearly positioned it that way.

The takeaway: across a CEO, a partner, and two customers, the recurring theme is the same — governed business context, not raw model capability, is what unlocks enterprise adoption.

What is available now?

Availability is split. Databricks said Genie and Genie Ontology are generally available today. Genie Search, Ontology Manager, and the always-on version of Genie are being rolled out through preview and beta programmes.

That split matters for buyers. The always-on capability — the headline feature — is not yet broadly shipping, so enterprises evaluating it now will be working with preview programmes rather than a finished product.

The broader stakes are easy to state. Enterprise AI agents have so far stalled at the pilot stage in many large organisations, blocked less by model quality than by questions of access, permission, and trust. Databricks' bet is that an agent which knows what the company means by "revenue," respects who can see what, and audits its own actions can cross that line — and the preview rollouts over the coming months will show whether customers agree.