Databricks Genie Agents Tackle Complex Multi-Step Tasks

Databricks unveiled an updated version of its AI agent, Genie, capable of unsupervised multi-step job completion. This enhancement is powered by an ontology designed to provide Genie with context from customer data on the Databricks platform. Leaders can now assign diverse tasks to the agent, which will execute them without supervision.

Databricks CEO Ali Ghodsi showcased Genie's internal application. He demonstrated how it analyzed negative feedback from an employee satisfaction survey and helped evaluate proposed solutions. The agent also summarized a poorly received sales pitch, identifying the prospect's required features. Additionally, Genie prepped Ghodsi for a podcast by reviewing past appearances and generating likely questions.

Ghodsi highlighted Genie's role in providing candid feedback, especially for leaders in large organizations exceeding Dunbar's number of 150. He emphasized that at Databricks' scale of 12,000 employees, direct feedback is challenging, making AI agents like Genie crucial for uncovering unaddressed issues. He believes Genie can offer significant value to companies with over 150 employees.

Databricks is now promoting Genie as an always-on agent for business leaders, with Ghodsi himself using it personally. He likens Genie's impact on the business side to that of AI and agents for programmers. To address potential AI cost increases, Ghodsi mentioned Databricks' Unity Gateway product, which allows for usage limits and the integration of less expensive open-weight models.