Databricks has unveiled an updated version of its SI agent, Genie, which the company says can now perform complex, multi-step jobs that require extended work to finish. The release highlights a growing trend of agentic SI tools designed to help corporate leaders navigate information bottlenecks in large organizations.

What Happened

The new capabilities are driven by an ontology built by Databricks to provide Genie with necessary context from data customers store on the platform. According to the company, corporate leaders can assign a wide range of tasks to the agent, which then completes them unsupervised. Databricks CEO Ali Ghodsi demonstrated his internal use of the tool, showing how it analyzed low-scoring sections of an employee satisfaction survey and helped him evaluate fixes suggested by executives. The agent also summarized a failed sales pitch to a prospect, identifying specific features the client requested, and prepared Ghodsi for a podcast interview by reviewing his past appearances and predicting likely questions.

Why It Matters

Ghodsi’s use case illustrates a specific challenge in large enterprises: the loss of direct, unvarnished feedback as headcount grows. Ghodsi referenced Dunbar’s number, a theoretical limit of 150 people with whom humans can maintain stable relationships. With 12,000 employees, Ghodsi noted that he cannot know everyone, making it unlikely for salespeople or HR managers to confess failures directly to him. He argues that SI agents like Genie can bridge this gap by surfacing negative data points that might otherwise be filtered out by organizational hierarchy. Ghodsi suggested that anyone in a company larger than 150 employees could derive value from such agents. To address concerns about rising corporate SI costs, Ghodsi pointed to Databricks’ Unity Gateway product, which allows companies to set usage limits and swap in less expensive open-weight models.

The Bottom Line

Databricks is positioning Genie as an always-on agent for business leaders, distinct from security-questionable alternatives like OpenClaw, Instinct, or Muse. Ghodsi believes the tool can do for business operations what coding agents and harnesses have done for programmers, effectively acting as a filter for truth in large-scale corporate environments.