Databricks has moved Genie from a question-and-answer tool to an always-on AI coworker: autonomous agents that take assignments, run scheduled and multi-step work, and keep producing insights in the background, on governed business context. Why it matters: analytics is shifting from asking a chatbot questions to delegating work to agents that act while you are doing something else.
November 11, 2026
What changed?
Genie, Databricks’ business-intelligence product, will now accept assignments instead of only answering individual questions. A user can give it a task such as “every morning, tell me which regional sales reps are underperforming and why,” and Genie will run it on a schedule and deliver results proactively. It can also chain steps together: a result from one question becomes the input to the next, or results can be used to create a new curated data asset.
Databricks is shipping the pieces behind that shift as generally available: Genie Ontology, which codifies an organisation’s business context — metrics, definitions, synonyms and usage patterns — so different users get consistent answers; and Genie Search, a semantic search layer over data in Databricks SQL that can look across schemas, tables and columns, including unstructured data, while respecting governance.
Why agents, and why now?
The pitch is that a single prompt-and-response loop was never enough for real analytical work, which is usually a sequence: define the question, check the data, run the analysis, verify the result, write it up. Databricks says Genie can now take that sequence, not just its first step, and keep it running in the background.
That framing matters because every major vendor is making a version of the same claim, from Salesforce’s agentic layer to Microsoft Copilot. What distinguishes Genie’s pitch is that the agent work runs on top of the data platform where the governed data already lives, rather than on a separate application layer. If the governance story holds, Genie’s agents should not be able to reach data a user is not allowed to see; if it does not, an always-on agent becomes a new way to leak data at scale.
What to watch
The interesting test is not the demo — it is whether scheduled, multi-step agents produce answers a finance or sales team will trust without re-checking every step. Databricks’ answer is the Ontology: if the business context is codified once, agents should interpret questions the same way every time. That puts the burden on data teams to maintain that context well.
For Tested Intelligence readers, the practical takeaway is simpler: if you already run Databricks SQL, Genie just became a way to delegate recurring analytical work rather than a way to ask one-off questions. The question to ask in a pilot is not whether the agent can answer, but whether it can do so repeatedly, on schedule, without drifting from the agreed definitions.
Frequently asked questions
What is Genie Ontology?
A generally available layer in Databricks that codifies business context — metrics, definitions, synonyms and usage patterns — so Genie gives consistent answers across users and questions.
What is Genie Search?
A generally available semantic search layer over data in Databricks SQL that can search across schemas, tables and columns, including unstructured data, while respecting data governance.
What can Genie agents do that the old Genie could not?
Take assignments, run scheduled and multi-step work in the background, and proactively deliver results, rather than only answering individual questions in a chat.
Sources: Databricks (product announcement, 11 November 2026); The Register (coverage)