Databricks Unveils Genie AI Agent for Multi-Step Tasks

The updated Genie agent handles extended, multi-step jobs using a new ontology layer built on data customers already store on the Databricks platform.

Updated on Nov. 11, 2026 in Artificial Intelligence

Databricks Unveils Genie AI Agent for Multi-Step Tasks

Databricks unveiled an updated version of its Genie AI agent on Wednesday, capable of taking on multi-step jobs that require extended work. The core of the update is Genie Ontology, an ontology layer that maps what each table means, how its pieces connect, and which numbers matter for which questions.

Why it matters

The update extends Genie from answering individual data questions toward carrying out longer, unsupervised workflows for businesses that already run on the Databricks platform. Databricks argues the added context should make the agent less prone to fabricating answers.

Genie Ontology draws on context from information customers already store on the Databricks platform, with agent mode adding deeper reasoning and support for unstructured files in Unity Catalog volumes. Databricks has not published independent accuracy benchmarks, and the extent of the claimed reduction in hallucinations remains unverified.

The players

Databricks

Databricks is a San Francisco-based data and AI company whose platform combines data warehousing, machine learning and analytics, and which develops the Genie AI agent family.

The details

Agent mode lets files stored in Unity Catalog volumes feed into the agent's work, while Genie Agents combine inferred knowledge about user data with semantics users define themselves. Databricks says the enhancements are expected to significantly reduce hallucinations, though it has not quantified that reduction in the announcement.

Timeline

  1. June 2026: Databricks introduced Genie One and Genie Agents at its Data + AI Summit.

  2. November 9, 2026: Genie Ontology is scheduled for general availability.

  3. November 10, 2026: A Chrome extension for Genie enters beta.

  4. November 11, 2026: The Genie One desktop app for Windows enters beta.

The Tech Race

Genie Ontology builds on Unity Catalog, the governance layer Databricks uses to manage data and AI assets, and extends it into agent behavior. The move positions Databricks against rival enterprise data platforms competing to make AI agents reliable on proprietary corporate data.

Data teams using Databricks could soon hand longer, multi-step analysis tasks to Genie rather than querying tables one question at a time. Users should watch the beta releases and test the agent against their own data before relying on its outputs.

The takeaway

Enterprise AI agents become useful when they understand the meaning of a company's data, not just its structure. Teams evaluating these tools should verify outputs against known answers before trusting them with unsupervised work.

What happens next

Genie Ontology reaches general availability on November 9, 2026, with the Chrome extension beta following on November 10 and the Windows desktop app beta on November 11.

Further reading

For more on how companies are deploying autonomous AI tools, see Artificial Intelligence.