Databricks unveiled an updated version of its AI agent Genie on Wednesday that can now complete multistep jobs without supervision. The agent uses an ontology built on the Databricks platform to give it context from customer data.
CEO Ali Ghodsi demonstrated the agent's capabilities by sharing his own internal usage rather than a traditional product demo. Ghodsi showed how Genie analyzed underperforming sections of an employee satisfaction survey, rated executive-proposed fixes, summarized a failed sales pitch including the prospect's stated needs, and prepared him for a podcast interview by reviewing his previous appearances and predicting likely questions.
Ghodsi framed the agent's value for executives at scale. He noted that while CEOs at small startups maintain direct relationships with employees who share negative feedback openly, his 12,000-person company presents a different dynamic. "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, referencing Dunbar's number, the theoretical maximum for stable human relationships.
Ghodsi uses Genie in his personal life as well as at work. He positioned the agent as doing for business leaders what AI agents and harnesses are doing for programmers. On the cost risk of enterprise AI agent deployment, Ghodsi pointed to Databricks' Unity Gateway product, which allows companies to set usage limits and swap in less expensive open-weight models.