Databricks has updated its Genie AI agent to handle business tasks that require multiple steps and take time to complete. CEO Ali Ghodsi demonstrated how he uses the agent inside the company, where it helps him examine employee feedback, sales conversations and other information that might be difficult to surface through ordinary management channels.
From answers to extended tasks
Databricks says the updated Genie can work through a broad range of assignments without supervision. It uses an ontology built by Databricks to provide context from information customers store on the platform. That context is intended to help the agent connect a request to relevant business data and carry out work beyond a single response.
During his demonstration, Ghodsi showed his actual chats and charts rather than a prepared product example. One use was reviewing an employee satisfaction survey to identify areas where the company scored poorly. Genie also helped him assess proposed fixes from executives, including deciding which ideas to discard.
Another example involved a sales pitch to a prospective customer that had gone badly. Genie summarized the discussion, including features the prospect said it would need before buying. In that case, the agent helped turn an unsuccessful conversation into information Ghodsi could review and act on.
A channel for difficult feedback
Ghodsi also used Genie to prepare for a podcast interview. The agent reviewed his earlier appearances and developed likely questions. During the conversation with TechCrunch, he asked it for suggestions on what to say to the interviewer.
The uses share a larger purpose: finding information that may not reach a company leader directly. As organizations grow, employees can become less likely to bring problems or bad news straight to the CEO. A salesperson may hesitate to describe a failed call, for example, and an HR manager may not volunteer that a proposed morale initiative is unlikely to work.
Ghodsi connected this challenge to Dunbar’s number, which he described as 150: the theoretical maximum number of people with whom a person can maintain stable relationships. In a small startup, he said, a CEO can know everyone personally, making candid conversations easier. With 12,000 employees, that direct familiarity is no longer possible.
He said that people at companies with more than 150 employees could get value from AI agents such as Genie. The examples he showed suggest one possible role for the technology: organizing records and feedback so leaders can inspect uncomfortable information, even when no one has raised it with them face to face.
Genie as a business tool
Databricks is pitching Genie as an always-on agent for business leaders, in part because Ghodsi uses the product himself. That practice, often called dogfooding, means a company uses its own product internally. Ghodsi said he also uses Genie in his personal life, rather than alternatives he considers security-questionable, including OpenClaw, Instinct or Muse.
He sees a parallel with AI tools used by programmers. Agents and harnesses, which are software wrappers that let an AI model use tools, are changing how programmers work. Ghodsi believes Genie could bring similar capabilities to business tasks.
Wider use also raises a practical question: how to keep AI costs under control. Ghodsi pointed to Databricks’ Unity Gateway product, which he said lets companies set usage limits and switch to less expensive open-weight models.
The demonstration showed Genie being used to gather and summarize company information, surface weak spots and support preparation. Databricks’ broader pitch is that an agent with access to relevant business context can take on extended work and help leaders see issues they might otherwise miss. Its usefulness, as described by Ghodsi, depends on putting company information to work in ways that support candid review and follow-through.