What happened
Databricks on Wednesday released a revised version of Genie, its AI agent, which can now work through multistep assignments that take extended effort to complete, TechCrunch reported. According to the company, Genie gets its context from information customers keep on the Databricks platform, via an ontology Databricks built, and executives can give it a broad range of tasks to finish without supervision.
Chief executive Ali Ghodsi showed how he uses the agent inside the company, opening his own recent sessions in place of a scripted demo. In one, Genie examined the weakest areas of an employee satisfaction survey and helped him evaluate, and reject, remedies some executives had proposed. In another, it summarized a sales pitch that went badly, including features the prospect wanted before buying. It also prepared him for a podcast by reviewing past appearances.
Ghodsi pointed to Dunbar's number, the theory that people can sustain stable relationships with roughly 150 others, and said that at 12,000 employees he cannot know everyone or expect bad news to reach him unprompted. He said any organization with more than 150 staff could benefit from such agents. On spending, he said the Unity Gateway product lets companies set usage limits and switch to cheaper open-weight models.
Context
Ghodsi's internal use of Genie, known as dogfooding, helps explain why Databricks now promotes it as a continuously running assistant for executives. He said he relies on it outside work too, preferring it to options he sees as security risks, such as OpenClaw, Instinct or Muse. In his view, Genie could bring business teams the gains that AI, agents and tool-using harnesses are delivering for programmers.
ब्रीफ़िंग
टेक का पूरा सप्ताह, एक ईमेल में।
हर सप्ताह वे रिपोर्ट्स जो आपके समय के लायक हैं। कभी भी अनसब्सक्राइब करें।
