Databricks has unveiled a second generation of Genie, its enterprise AI assistant, that works on its own in the background instead of waiting to be asked. Employees give Genie a goal and it takes ownership of the work, delegating to sub-agents, tracking progress and reaching out when a person needs to make a decision. The catch: the always-on mode itself is not shipping yet — it is coming soon to preview and beta programs.
What the new Genie does
The new Genie is built around a simple shift: from answering questions to owning jobs. It can keep a quarterly forecast on track, monitor customer usage and flag meaningful shifts, or track revenue trends and deliver a weekly brief. Marketers can ask it to watch GEO performance and suggest content; customer success teams can ask it to monitor accounts.
Genie opens to a priority inbox of items that need attention, and each new alert arrives as its own unread email, with updates threaded together. Every email links back into the Genie app. The assistant runs on its compute in the background, on a schedule or when triggered, and only comes back to a person when a decision is needed.

The context layer underneath
Everything Genie does is grounded in Genie Ontology, Databricks’ layer that captures how a company actually works: its definitions, metrics, people and processes. The ontology learns from dashboards, queries, notebooks and external BI tools, and keeps itself current as the business changes. It is the reason Genie can answer from a company’s own numbers instead of generic knowledge.
Genie Search extends that context across the tools a company already uses, with more than 100 connectors to Google Drive, Jira, Confluence and the lakehouse. Content can be indexed in place, ingested through Lakeflow Connect or queried live through MCP, without moving or copying it. Agents get what they need in a single search instead of checking each app separately.
Governed by default
Databricks’ pitch is that Genie works inside the controls companies already have, not around them. Every action stays within Unity Catalog permissions and Unity Gateway approval rules and budgets, and every action is fully auditable. Admins decide which tools Genie can use and set spending caps, and Genie only sees what each person is allowed to see.
The governance claim matters because the failures of autonomous agents are usually permission failures. United Airlines just told AI agents they cannot book flights, drawing a line around what agents may do on other companies’ systems. Databricks is betting that enterprises will trust an agent that stays inside their own permissions — and can prove it did.
Our Read
Read plainly, this is Databricks moving Genie from a question-answerer to a co-worker. The announcement leads with autonomy, not accuracy: the company is selling an agent that knows the business and can be trusted with work, not just with answers. Agents have already run a DNA lab, so the question is no longer whether they can do real work.
The honest limit is the one in Databricks’ own availability note. Genie and Genie Ontology are generally available, but Genie Search, Ontology Manager and the always-on mode are only ‘coming soon’ to preview and beta, with no dates. The headline says Genie ‘is now’ always on; the fine print says the always-on part is not yet in customers’ hands.
Why It Matters
It matters because the agent race is moving from chat to ownership of work. Every major AI company is now selling agents that do jobs, not just answer questions — and the enterprise version of that race will be won on governance, not capability. Databricks’ 20,000 customers give Genie a distribution channel most agent startups can only dream of.
For buyers, the test is whether Genie knows their business well enough to be useful — and whether the audit trail is strong enough to trust it. The preview and beta period will answer both questions. Until then, the announcement is a statement of direction: Databricks believes the enterprise agent is a co-worker, not a chatbot.
Background
Databricks announced the second generation of Genie on stage at Data + AI Live, in a session between chief executive Ali Ghodsi and OpenAI president Greg Brockman. Genie and Genie Ontology are generally available now; the always-on mode, Genie Search and Ontology Manager follow in preview and beta.
Source: Databricks’ announcement.


0 Comments