AI WORLD HQ
If you have ever asked a data team, "What exactly counts as a sale this quarter?", you already know the problem Databricks is trying to solve. The answer lives in someone's notebook, an old dashboard, or a saved query, not in the database tables themselves. AI agents, which are programs that can answer questions and carry out tasks on their own, face the same puzzle.
What Databricks announced
Databricks is a company that provides a data platform, the software businesses use to store and analyze their information. Its Genie product lets people ask questions in plain language and get answers drawn from company data.
According to Databricks, the new work centers on Genie Ontology, which it describes as an agentic ontology. An ontology here means a shared map of what things mean inside a business, such as how "revenue" or "active customer" is defined. The company says agents can build this map from the work teams already do, while people review and govern it. Genie Ontology is now generally available, which means customers can use it in production.
Databricks also says it ran a new benchmark, a standardized test, built with the firm Turing. The test used business questions inside a live Databricks workspace, the shared environment where a team's data and tools live. The company says the test looked beyond tables and included notebooks, dashboards, and saved SQL queries (short database commands), because real answers often depend on those too.
Its reported results: Genie passed 91% of runs, compared with 45% and 58% for two other frontier agents, which are the most advanced AI agents available. Databricks also says Genie used less than half the tokens. A token is a small chunk of text, roughly a word or part of one, that an AI reads and writes. Fewer tokens generally means less work for the system. The company additionally reports that Genie finished three to four times faster.
These are Databricks's own numbers from its own test. They are useful signals, but they are not independent verification.
Genie One: the cowork idea
Databricks also introduced Genie One, which it calls the enterprise home for AI cowork. "Cowork" here means AI that helps you finish work rather than only answering questions.
The company describes several pieces:
- Genie Search: a search tool that looks through company documents, tickets, and wikis, and can also reach live information.
- Connections to outside tools: Databricks says Genie Search supports more than 100 MCP connectors. MCP (Model Context Protocol) is a standard way to give an AI access to apps such as Jira, so it can check a ticket's status or create one.
- Admin controls: a tool called Unity Gateway, which Databricks says gives administrators traces (records of what the agent did), policies for connections, and budget controls.
Databricks also says Genie can now take more autonomous action, meaning it can carry out steps across teams and tools with less back-and-forth.
Wrap-up
Databricks is betting that AI agents become more useful when they understand how a business thinks about its own data. Its reported test results and its new Genie One features point in that direction, though the numbers come from the company itself. The next step is simple: pick one recurring business question, find where its answer is defined today, and check whether that definition is written down anywhere your team can see it.