Databricks launches single-prompt Genie Agent creation

Databricks has introduced a streamlined creation process for its Genie Agents, enabling developers to spin up domain-specific AI assistants using a single prompt. Previously requiring manual configuration, these agents can now be generated instantly via Genie One or Genie Code. The system leverages the user's conversation alongside data governed by Unity Catalog to establish a functional starting point for automated workflows.

Rather than relying on constant prompt engineering to fix inaccurate answers, the new approach focuses on grounding the agents in existing corporate knowledge. Genie Agents can reason across structured databases as well as unstructured files like presentations, PDFs, Word documents, and images stored in Unity Catalog volumes. When a user queries the agent, it retrieves relevant files and cross-references them with governed tables, while strictly adhering to the user's individual access permissions.

For practitioners, Databricks recommends starting with a narrow, well-defined use case, such as tracking supply chain delays or investigating IT incidents. To ensure reliability, the platform includes built-in benchmarking tools. Developers can upload a set of test questions with expected answers to generate an objective accuracy score. This allows teams to measure performance improvements quantitatively as they expand the agent's capabilities to handle more complex, multi-step tasks.

Ultimately, this update shifts the developer's role from writing complex prompts to curating high-quality data sources. By utilizing Unity Catalog semantics like metric views and certified domains, organizations can quickly deploy reliable self-service tools. This transition ensures that AI assistants draw from verified financial tables or official product catalogs rather than guessing from unverified sources.