Key points
- Databricks unveiled an updated Genie agent on Wednesday that can handle multistep jobs needing extended work, which it says run unsupervised.
- Genie draws context from customer data on the Databricks platform through an ontology the company built.
- CEO Ali Ghodsi uses it internally for survey analysis, sales call summaries and interview prep at a company of 12,000 employees.
- Ghodsi says Unity Gateway lets companies set usage limits and swap in cheaper open-weight models to control agent costs.
Why it matters for marketers: Databricks is pitching long-running agents as a way for non-technical leaders to get candid analysis from company data, with cost controls built in, a template for marketing ops use cases.
Databricks on Wednesday released an updated version of Genie, its AI agent, which can now take on multistep jobs that require extended work. The company says leaders can assign a wide range of tasks and the agent will complete them without supervision. It relies on an ontology Databricks built to give Genie context from the information customers keep on the Databricks platform.
CEO Ali Ghodsi demonstrated the tool to TechCrunch using his own account rather than a staged demo. He had Genie analyze the weak spots in an employee satisfaction survey and rate fixes suggested by executives, summarize a sales pitch that went badly, and prepare him for a podcast by reviewing past appearances. Ghodsi framed the value as honest feedback. At 12,000 employees, he said, people are unlikely to bring a CEO bad news, and he cited Dunbar's number of 150 to argue any company above that size could benefit.
For marketers and content teams, the interesting part is the use case more than the product. Summarizing failed pitches, pulling out the features a prospect wanted, and testing internal ideas against data are tasks that sit in sales, marketing and communications. The agent works from data already stored in the platform, so results depend on how much of your customer and campaign information lives there.
Cost is the obvious concern with agents that run for long stretches. Ghodsi said Databricks' Unity Gateway lets companies cap usage and substitute less expensive open-weight models. Teams evaluating long-running agents should ask what controls exist for spend, and how the vendor defines unsupervised work before handing over anything customer-facing.
Sources
This brief is our own summary. Read the full reporting at the source.