Genie One: How Databricks Is Reshaping Its Data Assistant

Databricks has reworked Genie, moving it from a tool focused on answering questions about data toward one intended to help users act on it. This shift is packaged under Genie One, alongside two related developments, Genie Agents and Genie Ontology, that together change what Genie is scoped to do and how it arrives at its answers. This article outlines what each of these pieces consists of and how they relate to one another.

Genie One

Genie originated as a conversational analytics assistant within Databricks AI/BI, built around answering natural-language questions inside a defined data space. Genie One is described by Databricks as the next iteration of that assistant, framed internally as a "data-smart AI coworker", with the stated aim of extending Genie's role from answering questions to helping carry out the resulting actions.

To that end, Genie One connects to a broader set of systems than earlier versions did. Through Lakehouse Federation, Lakehouse Connect, and new two-way integrations with tools such as Gmail, Slack, and Teams, Genie One is designed to draw on and act across systems beyond the Lakehouse itself. Its feature set now includes scheduling and alerting, monitoring, document generation, custom skills, and support for custom MCP connections, capabilities that go beyond the query-and-answer pattern of earlier Genie spaces.

Databricks illustrates this with examples such as a sales leader having Genie assemble a daily briefing ahead of customer meetings by combining calendar, email, and Lakehouse data, or a general manager having Genie update an existing business review using current inventory figures and recent meeting transcripts.

Genie is also now embedded directly in Slack and Microsoft Teams, where it can be invoked via mention in a conversation, including in public channels and threads, rather than requiring a dedicated interface. Responses remain scoped to what each user is individually authorized to see. A mobile client is available on iOS and Android, and for organizations that already run their own AI agents, Databricks has introduced a Genie MCP App so those agents can call on Genie without a separate workflow.

Genie Agents

Genie Spaces, the curated, topic-scoped chat experiences that made up the earlier product, have reportedly been created more than a million times by Databricks customers. Databricks is now positioning an evolution of that concept, Genie Agents, as domain-specific agents built on the same underlying capabilities as Genie One: MCP connections, scheduled tasks, and document or artifact generation, applied to multi-step workflows that run with less ongoing supervision than a chat-based Genie Space required.

Where a Genie Space was generally grounded in structured data, Genie Agents can also draw on documents, files, and other unstructured sources. Databricks describes the creation process as similarly lightweight to before: an agent is described in natural language within Genie One or Genie Code, then scoped, benchmarked, and shared for others to use or adapt.

Genie Ontology

Both Genie One and Genie Agents draw on a new context layer, Genie Ontology, which extracts information from tables, queries, dashboards, pipelines, and connected applications and organizes it into a graph representing how an organization's data is structured and used, covering metric definitions, business terminology, custom calculations, and the relationships between them.

A notable design choice in Genie Ontology is how it ranks competing definitions of the same term or metric. Using an approach Databricks compares to PageRank, it weighs the source of a definition, the standing of whoever authored it, how frequently it's referenced, its proximity to certified assets, and how recent it is, then answers using the highest-weighted source while still enforcing each user's existing permissions. Databricks states that in its internal benchmarking on enterprise data analysis tasks, this approach improved agent performance on more complex questions, though this has not been independently verified.

Governance

Genie One's governance model builds on the access controls already present in the platform rather than introducing a separate system. Permissions are enforced by default through source-native ACLs or Unity Catalog, and MCP connections, tools, and associated costs are managed through the Unity AI Gateway, giving administrators one place to oversee usage.

Conclusion

Set against the earlier version of Genie, the change is less about individual features and more about scope. The original Genie Spaces model was built around answering a question within a single, curated space, a data professional set up the space, and a business user asked it something and got a result back. Genie One, Genie Agents, and Genie Ontology extend that same underlying idea across a wider set of systems and a wider set of tasks: connecting to tools beyond the Lakehouse, acting on information rather than only surfacing it, and resolving competing definitions automatically instead of relying on manually curated context. Whether that broader scope holds up to the same standard of reliability that the narrower, curated Genie Spaces model was built to guarantee is likely to become clearer as organizations put it into production use.