Databricks on Wednesday launched an enhanced edition of its AI agent Genie, now capable of handling multistep assignments that require extended effort to complete. Chief Executive Ali Ghodsi demonstrated the technology by revealing how he personally deploys it to uncover uncomfortable truths about his 12,000-employee company — including disappointing employee survey results, failed sales presentations, and executive proposals likely to backfire. Rather than showing a polished product demonstration, Ghodsi opened his actual Genie interface and walked through recent conversations and visualizations he'd generated.
The updated agent operates through an ontology Databricks constructed to supply Genie with necessary context drawn from information customers maintain on the Databricks platform. Company executives can now delegate a broad spectrum of assignments, and the agent will finish them without supervision, according to the company. Ghodsi's demonstrations included how the tool dissected portions of an employee satisfaction survey where Databricks performed poorly, helped him evaluate and reject remedies that certain executives had proposed for those weak areas, summarized a major sales pitch to a potential customer that went badly — including capabilities the prospect demanded before purchasing — and prepared him for a podcast appearance by analyzing his earlier interviews and crafting probable questions. During the interview itself, he even requested the agent's guidance on what to tell the journalist.
The most striking aspect was Ghodsi's reliance on Genie to surface information he's unlikely to receive as the top leader. It served as a channel for unfiltered reality about problems requiring his attention. "Have you heard of Dunbar's number? The magic number is 150," Ghodsi said when asked about this pattern, referring to British anthropologist Robin Dunbar's theoretical ceiling for how many people humans can sustain stable relationships with. In a tiny startup, a CEO maintains personal connections with everyone and receives negative information directly, he explained. With 12,000 staff members, Ghodsi acknowledged he can't possibly know everyone. At that scale, a salesperson won't admit a catastrophic sales call to the CEO, and an HR manager won't confess their employee satisfaction idea will likely flop. "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 internal use of Genie — often called dogfooding — is one driver behind Databricks now marketing it as an always-on agent for business leaders. He indicated he also applies it in his personal life rather than security-questionable alternatives. He believes Genie can deliver for the business domain what AI, agents, and harnesses — the software wrapping around an AI model that enables it to employ tools — are accomplishing for programmers. Regarding the possibility that agents will escalate corporate AI expenses, Ghodsi offered a solution: costs can be managed through Databricks' Unity Gateway product, which permits companies to establish usage caps and substitute less costly open-weight models. The structural challenge this technology addresses isn't technical but organizational — leaders at scale systematically lose access to ground truth about what's failing, and automated analysis can restore that visibility where human hierarchies can't. For enterprises already invested in the Databricks ecosystem, the agent's integration with existing data repositories may lower the friction to adoption that standalone tools would face.