Databricks updates its Genie AI agent for multi-step work

In short: Databricks has released an updated version of its Genie AI agent, and the company says it can now handle longer, multi-step tasks with less hands-on oversight.

What happened

Databricks announced an updated version of Genie, an AI agent (a helper that can take a goal and carry out a series of steps) designed for business work. The company says Genie can now complete jobs that take longer to finish and involve multiple steps.

Databricks says Genie does this by using an “ontology,” which is basically a structured map of how a company’s information is organized (like a labeled filing cabinet). That map is built from data customers store inside the Databricks platform, so the agent has more context when it answers questions and creates reports.

Databricks CEO Ali Ghodsi showed how he uses Genie inside the company. He used it to review weak spots in an employee satisfaction survey, and to rate and reject proposed fixes from executives. He also used it to summarize a sales pitch that went badly, including what a potential customer said it would need before buying.

Why it matters

As companies get bigger, leaders often stop hearing about day-to-day problems, because people do not want to bring bad news to the CEO. Ghodsi pointed to “Dunbar’s number,” the idea that most people can only keep stable relationships with about 150 people, to explain why this happens at scale.

Tools like Genie are being pitched as a way to surface that missing feedback, and to do routine prep work like summaries and interview questions. Databricks also says companies can control costs using its Unity Gateway, which can set usage limits and allow swapping in cheaper AI models.

Source: TechCrunch AI