Databricks Genie tells the CEO the truth. Your reporting deck doesn't

Picture this: your boss has already read the campaign numbers before you send your reporting deck. Not the slides. The raw data, summarized by an agent with no agenda. That is now a product Databricks is selling.

What Databricks showed

• The agent: On November 11, Databricks unveiled a new version of Genie, pitched as an always-on agent for business leaders. It runs multistep jobs, hands work to sub-agents and pings you when it needs input. Underneath sits an ontology that gives it context from the customer's data on Databricks.
• Status: per the announcement, Genie and Genie Ontology are generally available; search across Google Drive, Jira and Confluence and the always-on mode are in preview or beta. No pricing was given.
• The CEO's examples: according to TechCrunch, Ghodsi has used Genie to analyze the weak spots of an employee survey, rate fixes proposed by his executives and summarize a sales pitch that went badly, including the features the prospect said it needed.

His reasoning: past roughly 150 people, Dunbar's number, a CEO stops hearing things directly. Databricks has about 12,000.

This isn't a productivity tool. It's a power shift

Every organization has filters. Not out of malice, but because every layer condenses, weights and explains. In marketing, the monthly report is exactly that filter: we pick the metric, the comparison period, the story. When the CEO asks an agent sitting in the same data, the deck loses control of the narrative. What's left is one question: is the data the agent reads actually right?

That's where it gets uncomfortable. My rule of thumb: AI doesn't fix broken ways of working, it widens the cracks in the foundation. If "lead" has three definitions in your CRM and your campaign tags are half-maintained, the agent hands the CEO an honest answer built on bad data. Faster and more convincingly than any deck.

The use case marketing gets instantly

The most interesting customer example comes from Berlin. HelloFresh says in the announcement that the time it takes to identify winning ads on Meta dropped by 75%. That's a customer statement in a vendor launch, not an independent measurement. But it shows the direction: analysis is no longer the bottleneck. Judgment is. Which ad scales, which message fits the brand, what the budget can carry. That stays in-house, with a human.

The catch

"Unfiltered" doesn't mean "true." An agent only knows what's in the systems, not what gets said in the hallway. On the plus side, the reporting says Genie runs inside the customer's Databricks environment, respects existing access rights and logs every action. The agent sees no more than the person asking.

What I'd do now

• → Write down your five most important marketing metrics in one place, with an owner. Those definitions are the first thing any agent reads.
• → Rebuild your reporting: fewer number slides, more interpretation and decision framing. Soon anyone can pull the numbers themselves. Not the judgment.
• → Test it once: ask your data tool the question leadership would ask and check whether the answer matches your deck.

Bottom line: When the boss can query the data directly, the asset isn't the prettiest deck anymore, it's the cleanest definition.