Databricks introduced an upgraded version of its AI agent Genie on Wednesday, which is now capable of handling multi-step tasks that require extended effort to complete. The agent achieves this through an ontology Databricks developed to supply Genie with the necessary context from customer data stored on the Databricks platform. According to the company, business leaders can now delegate a broad range of tasks to the agent, and it will carry them out without supervision.

To demonstrate this capability, Databricks CEO Ali Ghodsi walked me through his own internal usage rather than giving a standard product demonstration. He opened his actual agent and guided me through his most recent conversations and charts. He showed how it examined the portions of an employee satisfaction survey where the company received poor scores, and how it assisted him in evaluating—and rejecting—remedies for those low scores that some executives had proposed. He demonstrated how it condensed a major sales presentation to a prospect that went badly—including the features the prospect indicated were necessary before making a purchase. It also prepared him for a podcast interview by going over his earlier appearances and generating probable questions. During our conversation, he even had it advise him on what he should say to me. (And its recommendation on that was actually quite good.)

What stood out was that Ghodsi was employing Genie to uncover information he's now unlikely to hear as the top executive. It served as a source of unfiltered truth about problems that needed his attention. "Have you heard of Dunbar's number? The magic number is 150," he remarked when I brought this up. Named after British anthropologist Robin Dunbar, it represents in theory the upper limit of people with whom humans can sustain stable relationships. In a small startup, a CEO knows everyone personally, has direct relationships with them, and they feel comfortable sharing negative information openly. With 12,000 employees, Ghodsi readily acknowledges he cannot know everyone. At that scale, a salesperson is unlikely to admit a disastrous sales call to the CEO. An HR manager is unlikely to confess that their plan for boosting employee satisfaction will probably fail. "Anyone who has more than 150 employees—which is a lot of companies—a lot of people can get value out of this," he said regarding AI agents broadly and his product Genie specifically.

Ghodsi's personal use of Genie—often referred to as dogfooding, derived from the phrase "eating your own dogfood"—is one reason Databricks is now marketing it as an always-on agent for business leaders. He mentions that he also uses it in his personal life, rather than a security-questionable alternative like OpenClaw or Instinct or Muse. He believes Genie can accomplish for the business side what AI, agents, and harnesses (the software that surrounds an AI model and enables it to use tools) are achieving for programmers. Regarding the risk that agents will drive up corporate AI expenses, Ghodsi has a solution for that as well. He stated that costs can be managed through Databricks' Unity Gateway product, which enables companies to establish usage limits and to substitute less expensive open-weight models.