Overview
- Databricks announced an updated version of its AI agent Genie that the company says can execute multi‑step, long‑running tasks using context drawn from customer data stored on its platform.
- The agent uses a Databricks‑built ontology, which maps and labels customer data so the model can understand context, plan multi‑stage actions, and carry tasks to completion.
- CEO Ali Ghodsi publicly demonstrated internal use, showing Genie analyze weak spots in an employee survey, evaluate and discard proposed fixes, summarize a failed sales pitch, and prepare interview talking points.
- Databricks said customers can limit costs and swap in cheaper open‑weight models through its Unity Gateway product to control compute and model spending for always‑on agents.
- The company is pitching Genie as an executive aide for leaders who lack direct contact across large organizations, but the capabilities and security claims are based on company demonstrations and need independent validation.