Although Databricks already had a product called Genie, this tool has been revamped to such an extent that it warrants an entirely new narrative. Genie, as an “autonomous AI colleague,” is constantly on the job, automatically completing tasks, with humans serving merely as supervisors.
Genie’s reliability still depends on Genie Ontology, which maps the business context as relational data. The tool leverages existing policies but handles data much more decisively than previous AI tools.
Genie can be accessed through a variety of channels: Genie One, email, Slack, and Teams. Databricks emphasizes the versatility of the tooling, similar to the commercial narratives of the AI labs Anthropic and OpenAI, but with the added benefit of deep integration with Databricks customers’ own business data. Incidentally, Genie can also be accessed through external LLMs such as Claude and ChatGPT if a chatbot interface is the preferred method of communication.
Proactive approach
The new Genie operates without the user having to explicitly request it. The agent tracks the progress of processes and requests missing input where necessary. It even takes action as soon as new information comes in. Users set the boundaries and provide approval where necessary. It may therefore take some getting used to what Genie is capable of, with a learning curve involved in truly leveraging that proactivity.
An example from Databricks is that Genie can check incoming insurance claims for potential fraud. It can then send a file to an insurer when information in a document falls outside the guidelines. Jobs run according to a schedule or respond to triggers, such as a declining metric. Genie also recognizes recurring tasks, such as a weekly update. It proactively offers to take over an action like this. Users interact with Genie via Genie One, email, Slack, or Microsoft Teams.
Ontology as the foundation
An agent needs not only context but also as much relevant information as possible. Genie Ontology, which has been available for some time, therefore maps out the company: definitions, metrics, relationships, and organizational knowledge. Agents largely build this themselves using existing dashboards, queries, and notebooks. Teams define leading definitions in Unity Catalog, after which the OntoRank algorithm prioritizes them.
According to Databricks, this approach is already paying off. In a benchmark against Turing, Genie provided the correct answer in 91 percent of task executions. That’s 1.6 times as often as the strongest frontier agent, while Genie consumed less than half the tokens.
Genie Search combines an index of apps such as SharePoint, Jira, and Workday with live searches via MCP. The access permissions of each source remain in effect. Through plugins, the same context is also available in ChatGPT, Claude, Copilot, and Grok Bot. In this way, Databricks is working on interoperability and model-agnostic tooling, which is important for a data platform with these agent-based capabilities.
Control over actions and costs
With Unity Gateway, admins manage access to connectors, MCP tools, and approval requirements. Every action is logged. Budgets and spending caps can be set per workspace or per user.
Competitor Snowflake takes a similar approach and offers a direct equivalent to Databricks Genie. Snowflake positions Snowflake Intelligence and Cortex Code as a control layer for agentic AI, including a Business Ontology and AI Gateway. Snowflake’s CoWork-Bots also perform recurring tasks autonomously.