Databricks turns Genie into an AI agent for executives

What Ghodsi asks Genie to do

Instead of a prepared demo, Ghodsi opened his work agent and showed recent conversations and charts. Genie analyzed employee satisfaction survey results and identified areas where the company received low scores. It then helped Ghodsi evaluate measures proposed by executives and reject those he considered unsuitable.

The agent also summarized a failed presentation to a prospective customer, including the customer’s product requirements before purchase. Ahead of a podcast interview, Genie studied Ghodsi’s past appearances and prepared likely questions. During a conversation with a journalist, the agent even suggested what the CEO should say.

Feedback for the CEO

Ghodsi uses Genie to find information about problems that might not reach him directly. He connected this to Dunbar’s number—the theoretical limit of about 150 people with whom a person can maintain stable relationships. The term is named after British anthropologist Robin Dunbar.

At a small company, the CEO may know every employee personally, making it easier for people to share bad news. But Databricks has 12,000 employees. Ghodsi acknowledges that he cannot know all of them: a salesperson is unlikely to tell him about a failed call, and an HR manager might not tell him that their idea for improving employee satisfaction probably will not work.

Ghodsi says AI agents can be useful at companies with more than 150 employees. He uses Genie regularly at Databricks—a practice known as “eating your own dog food,” or testing your product in real work. The agent also helps him in his personal life. As alternatives, Ghodsi mentioned OpenClaw, Instinct and Muse, expressing doubts about their safety.

Controlling costs

Ghodsi believes Genie could benefit businesses in a way comparable to how AI, agents and software wrappers around models already benefit programmers. These wrappers let models use tools.

In response to concerns that agents will increase companies’ AI spending, Ghodsi pointed to Databricks Unity Gateway. The product lets companies set usage limits and switch to less expensive open-weight models.

Dataist’s take: In the scenario described, Genie gives executives an additional channel for feedback and a way to analyze work data. But the examples are based on Ghodsi’s own experience, while the company’s cost controls rely on Databricks Unity Gateway.