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

Summary

• Databricks unveiled an updated version of its AI agent Genie capable of performing multi-step jobs that require extended time to complete.
• The agent utilizes a custom ontology to access context from customer data stored on the Databricks platform.
• Corporate leaders can assign a wide range of tasks to the agent for unsupervised completion.
• CEO Ali Ghodsi demonstrated internal usage by having the agent analyze employee satisfaction survey data, rate potential fixes, summarize a failed sales pitch, and prepare for a podcast interview.
• The agent provided real-time advice to Ghodsi during a conversation with a reporter.
• Ghodsi uses the tool to access unvarnished feedback that is unlikely to reach him directly due to Dunbar's number limiting stable relationships to 150 people.
• Databricks pitches Genie as an always-on agent for business leaders with over 150 employees.
• Ghodsi uses Genie for personal tasks instead of security-questionable alternatives like OpenClaw, Instinct, or Muse.
• The agent aims to replicate the capabilities of AI tools for programmers on the business side.
• Databricks states that corporate AI costs can be controlled through the Unity Gateway product, which allows for usage limits and the use of less expensive open-weight models.