Databricks turns Genie into an always-on agent for leaders

By Grace Kim

What changed in Genie?

Databricks unveiled an updated Genie AI agent on Wednesday that can perform multi-step jobs unsupervised, and CEO Ali Ghodsi said the company now has 12,000 employees.

Databricks says corporate leaders can assign the agent a wide range of tasks, including work that takes extended time to finish, and it will complete them unsupervised.

An ontology built by Databricks gives Genie context from information customers store on the platform. The company now pitches Genie to corporate leaders as an always-on agent for business.

The practical change is the ability to carry out extended, multi-step assignments with company context.

How does Ghodsi use it?

Ghodsi did not stage a conventional product demo for a reporter. He fired up his actual agent and walked through his most recent chats and charts, showing how the tool handles management work inside Databricks.

According to Ghodsi, Genie analyzed portions of an employee satisfaction survey where Databricks did not score well. It also helped him rate and discard proposed fixes for those low scores that some executives had suggested.

According to Ghodsi, the agent summarized a large sales pitch to a prospect that had gone poorly. The summary included the features the prospect said it needed before buying.

For a podcast interview, Genie reviewed Ghodsi's previous appearances and developed likely questions, he said. During the demonstration, he also asked the agent what he should say to the interviewer, and it offered advice.

Genie's practical value here is its ability to review options and prepare a CEO for a public appearance from company data.

Can Genie deliver the unvarnished truth?

Ghodsi said he uses Genie to find information he is unlikely to hear as Databricks' big boss, including what is not going well and needs his attention.

In Ghodsi's account, a salesperson is unlikely to confess a disastrous sales call, and an HR manager is unlikely to admit that an idea for increasing employee satisfaction will probably fail.

“Have you heard of Dunbar’s number? The magic number is 150,” he said. The concept, named for British anthropologist Robin Dunbar, theoretically represents the maximum number of people with whom humans can maintain stable relationships.

In a small startup, the CEO knows everyone, has direct relationships with employees, and creates enough safety for them to deliver negative information straight up, Ghodsi said.

At 12,000 employees, Ghodsi acknowledged that he cannot know everyone.

“Anyone who has more than 150 employees, which is a lot of companies, a lot of people can get value out of this,” Ghodsi said of AI agents generally and Genie specifically.

At enterprise scale, Genie offers leaders a route to weak signals without relying only on subordinates to volunteer them.

Why does Databricks lean on its CEO?

Ghodsi's internal use of Genie is what people commonly call dogfooding, from the phrase eating your own dogfood. Databricks cites that CEO-level use as one reason it now positions Genie as an always-on agent for corporate leaders.

He also uses Genie in his personal life rather than a security-questionable alternative like OpenClaw, Instinct or Muse, he said.

The CEO's own routine is therefore part of the product's enterprise pitch.

Can companies control the bill?

Ghodsi believes Genie can do for the business side what AI, agents and harnesses are doing for programmers. He described harnesses as software that wraps around an AI model and lets it use tools.

He also addressed the risk that agents will ratchet up corporate AI costs. Companies can control those costs through Databricks' Unity Gateway product, he said. Unity Gateway lets companies set usage limits and swap in less expensive open-weight models.

For Databricks, the next test is whether customers can add unsupervised work while Unity Gateway keeps model spending bounded.