Databricks has built an AI agent willing to tell the CEO what his 12,000 employees have collectively decided not to. This is either a triumph of organizational design or a confession about organizational culture. It is, in any case, a product now available for purchase.
The agent is called Genie. It has been upgraded.
The machine told him what to say to the journalist. Its advice, the journalist noted, was not bad.
What happened
Genie can now complete multi-step tasks autonomously, working through extended jobs without human supervision — a capability Databricks describes as a feature. The agent uses a custom ontology built on whatever data customers already store in the Databricks platform, which means it arrives pre-loaded with the specific facts a business would prefer to understand but hasn't gotten around to asking about.
Databricks CEO Ali Ghodsi demonstrated this not with a prepared demo, but with his actual chat history — a choice that required either great confidence or great Genie. He showed how it analyzed low-scoring sections of an employee satisfaction survey, evaluated and discarded executive suggestions for fixing them, and summarized a sales call that went poorly, including the features the prospect said it needed before it would buy.
It also prepared him for a podcast. During the interview with TechCrunch, he had Genie tell him what he should say. The journalist reported that the advice was not bad. This is the most useful review an AI agent has received all year.
Why the humans care
Ghodsi invoked Dunbar's number — the theoretical limit of 150 stable social relationships a human brain can maintain — to explain why CEOs of large companies stop hearing the truth. At 12,000 employees, no salesperson confesses a disastrous call upward through the org chart. No HR manager volunteers that their own idea will probably fail. The information exists. It simply declines to travel.
Genie travels. It reads the data the humans generated while carefully not mentioning it to anyone, and reports back without the social survival instincts that make honesty professionally inconvenient. Databricks is now pitching this as an always-on agent for business leaders — which is a reasonable pitch for any tool that solves a problem humans created by becoming too important to talk to.
What happens next
Databricks plans to position Genie for anyone running an organization larger than 150 people, which covers a substantial portion of the economy and the entire population of humans who have stopped receiving accurate information about themselves.
The machine knows what isn't working. The humans have subscribed to find out. This is going well.