The key facts

  • Genie now performs multi-step tasks unsupervised, relying on customers' data ontology
  • Ghodsi uses the agent to analyze the company's weak spots and prepare for interviews
  • Spending on agents can be controlled via Unity Gateway

What Databricks unveiled

Databricks showed an updated version of its AI agent Genie. The main change: the agent can now perform multi-step tasks that require lengthy work and carry them through to completion unsupervised. This is achieved through an ontology the company built so that Genie gets the necessary context from data stored by customers on the Databricks platform. According to the company, executives can assign the agent a wide range of tasks, and it will complete them on its own.

How the CEO himself uses the agent

Instead of a standard demo, Databricks CEO Ali Ghodsi showed his real agent at work: recent chats and charts. Genie analyzed the sections of an employee satisfaction survey where the company scored low, helped evaluate and filter out fixes proposed by top managers. The agent also summarized a failed major pitch to a potential customer, including the features the customer listed as needed for a purchase, and prepared Ghodsi for a podcast interview by studying his past appearances and drawing up likely questions. During the conversation with the journalist, Ghodsi simultaneously asked the agent what he should say, and, in the interviewer's assessment, the advice was 'not bad'.

'Unvarnished truth' and the Dunbar number

As TechCrunch observed, Ghodsi uses Genie to learn things he, as the company's top executive, would hardly hear directly: the agent has become for him a source of unvarnished information about problems. Ghodsi explained this through the 'Dunbar number' — 150, the theoretical maximum of stable social connections a person can have, named after anthropologist Robin Dunbar. In a startup, the CEO knows everyone personally, and employees report negatives to him directly. With 12,000 employees, Ghodsi admits, he cannot know everyone: a salesperson won't admit to the CEO that a call went badly, and an HR manager won't admit that their idea for improving satisfaction will likely fail. 'Any company with more than 150 employees — and there are many such — can benefit from this,' he said about AI agents in general and Genie in particular.

Strategy and cost control

The CEO's personal use of Genie (what the industry calls dogfooding) is one of the reasons Databricks now promotes the agent as a constantly working tool for business leaders. Ghodsi added that he also uses it in his personal life, instead of alternatives that raise security questions such as OpenClaw, Instinct, or Muse. In his view, Genie can do for business tasks what AI, agents, and harness scaffolding do for programmers. To the risk of rising corporate spending on agents, he responds with the Unity Gateway product: it allows setting usage limits and switching to cheaper open-weight models.