Databricks this week introduced a significantly upgraded version of its AI agent, Genie, expanding its capabilities to handle multi-step tasks that require sustained effort over time. The update allows the agent to work through complex assignments autonomously, drawing on an ontology the company built specifically to give Genie the contextual grounding it needs to interpret and act on data stored within the Databricks platform.

According to the company, business leaders can now hand off a broad range of tasks to Genie and expect them to be completed without ongoing human supervision. To demonstrate this in practice, Databricks CEO Ali Ghodsi opted against a rehearsed product demonstration, instead opening up his own live instance of the agent and walking through actual recent conversations and outputs he had generated for himself.

A CEO Using His Own Product to Get Honest Answers

Among the examples Ghodsi shared was how Genie helped him dig into the weaker results from an internal employee satisfaction survey, surfacing the areas where the company had underperformed. Beyond simply identifying the gaps, the agent also helped him evaluate - and ultimately reject - proposed remedies that some executives had put forward to address those low scores.

In another case, Genie summarized the details of a sales pitch to a prospective client that had not gone well, including a breakdown of the specific product features the prospect said it required before it would consider purchasing. The agent also helped Ghodsi prepare for a podcast appearance by reviewing his previous interviews and generating a list of questions he was likely to face.

During the conversation, Ghodsi even had Genie weigh in on what he should say in the interview itself - and by his own admission, the guidance it offered was not far off the mark.

What stood out in all of these examples was a common thread: Ghodsi was using the agent to surface exactly the kind of candid, uncomfortable information that rarely makes its way to a chief executive through normal organizational channels. Genie was functioning, in effect, as a source of unfiltered truth about what was not working inside the company.

The Dunbar's Number Problem at Scale

When asked about this dynamic, Ghodsi pointed to a well-known concept from behavioral science. "Have you heard of Dunbar's number? The magic number is 150," he said. The figure comes from the work of British anthropologist Robin Dunbar, who proposed that human beings can realistically maintain stable social relationships with no more than roughly 150 people at any given time.

In a small startup, a CEO tends to know every employee personally. Those relationships are direct, and people generally feel comfortable sharing bad news or honest assessments without fear of repercussion. But Databricks now employs around 12,000 people, and Ghodsi is candid about what that means for the flow of information reaching him.

At that scale, a salesperson who had a disastrous client call is unlikely to bring that news directly to the CEO. An HR manager whose idea for improving workplace satisfaction is probably going to fall flat is not going to volunteer that assessment upward through the organization. The information exists somewhere in the company, but the organizational layers between frontline employees and the executive suite act as a filter, smoothing over inconvenient truths before they arrive at the top.

Ghodsi sees AI agents as a way to cut through that filter. "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, speaking about AI agents broadly and about Genie in particular.

Dogfooding as a Product Strategy

The fact that Ghodsi uses Genie regularly in his own work - a practice commonly referred to in the technology industry as dogfooding, derived from the phrase "eating your own dog food" - is a central part of why Databricks is now positioning the agent as a persistent, always-on tool for business executives.

He also noted that he uses Genie in his personal life, citing it as a more secure alternative to other AI tools he did not feel comfortable trusting with sensitive information. The implication was that for someone in his position, data privacy is a meaningful consideration when choosing which AI products to rely on day to day.

The Broader Vision and Cost Controls

Ghodsi's broader argument is that Genie can replicate for business operations what AI agents and the software frameworks built around AI models are already doing for software developers. Just as programmers have increasingly offloaded repetitive and time-consuming coding tasks to AI-assisted tools, he believes enterprise leaders can do the same with the operational and analytical work that currently consumes significant executive attention.

One concern that often accompanies broader enterprise AI adoption is the potential for costs to spiral as usage increases. Ghodsi offered a specific answer to that challenge, pointing to Databricks' Unity Catalog product as a mechanism for keeping spending in check. The platform allows organizations to set defined usage limits and to substitute less expensive open-weight AI models in place of more costly alternatives where appropriate, giving companies a practical lever for managing their overall AI expenditure without sacrificing capability across the board.