Databricks has launched ai_decide, a native AI function that turns unstructured text into structured decisions in a fraction of a second.

Now in beta, ai_decide is powered by TypeSafe AI’s Jev decision model and is designed for high-volume judgment calls: routing prompts to the right model, tagging customer feedback, evaluating agent outputs and embedding decisions directly in application logic. It is reachable through SQL for batch scoring and a REST API for real-time calls.

Unlike a generative model, ai_decide produces values to be consumed, not prose to be interpreted. Outputs are structured: probabilities, a choice from named criteria or a position on an ordered scale.

That difference changes the engineering around it. There is no output parsing, no retry logic for malformed responses, and evaluation compares scores against labels instead of judging prose. Databricks says it runs below the cost of a standard LLM call doing comparable classification work.

Why it matters: for teams drowning in guardrails around model output, a fast decision model is a meaningful simplification.