Databricks’ new AI agent tells CEO Ali Ghodsi the unvarnished truth

Databricks on Wednesday unveiled an updated version of its AI agent Genie, which can now carry out multi-step jobs that normally take extended effort. The company says the agent can complete assigned tasks unsupervised, drawing on an ontology Databricks built so Genie understands the context of information customers store on its platform.

Rather than a canned product demo, CEO Ali Ghodsi walked a TechCrunch reporter through his actual agent and his most recent chats and charts. He showed Genie analyzing the parts of an employee satisfaction survey where Databricks scored poorly, then helping him evaluate — and discard — fixes proposed by executives for those low scores. He also used it to summarize a poorly received sales pitch to a prospect, including the features the prospect said it required before buying.

Genie also prepped Ghodsi for a podcast interview by reviewing his previous appearances and forecasting likely questions; while talking with the reporter, he had it suggest what he should say in the conversation. The common thread, the article notes, is that Ghodsi used Genie to surface information he is unlikely to hear directly as the top executive — a source of unvarnished truth about what is going wrong.

Ghodsi framed the rationale around Dunbar's number, the theoretical maximum of roughly 150 stable relationships a person can maintain, named for British anthropologist Robin Dunbar. At Databricks' current size of 12,000 employees, he says he cannot know everyone, and staff are unlikely to volunteer bad news directly — a salesperson won't confess a disastrous call to the CEO, nor will an HR manager admit an employee-satisfaction idea is likely to fail. "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.

Ghodsi's own use of Genie — dogfooding, from the phrase about eating your own dog food — is a key reason Databricks now pitches it as an always-on agent for business leaders. He argues Genie can do for the business side what AI, agents and harnesses (the software wrapping a model so it can use tools) are doing for programmers. He says he also relies on it in his personal life rather than security-questionable alternatives he named as OpenClaw, Instinct or Muse. On the concern that agents will drive up corporate AI costs, he pointed to Databricks' Unity Gateway product, which lets companies set usage limits and swap in cheaper open-weight models.

What to watch: whether business leaders adopt unsupervised multi-step agents as broadly as developers have adopted coding agents, and whether cost controls like usage caps and open-weight model swaps keep pace with agent usage. Open questions remain around how much verification executives will demand before acting on an agent's unfiltered assessments, and whether the ontology-based approach generalizes beyond Databricks' own platform data.