Introducing ai_decide: make fast decisions on your governed data

Databricks has introduced ai_decide, a native AI Function designed to make rapid, cost-effective decisions over governed data. The function addresses common enterprise workflows such as categorizing support tickets, determining document review needs, or selecting appropriate models for prompts. Unlike LLMs, which generate text and incur higher latency and costs at scale, ai_decide evaluates questions against text in milliseconds, returning probabilities, choices, or scores without generating additional content.

ai_decide is powered by a decision model optimized for speed, reducing both latency and expense compared to LLMs for similar tasks. It is now available in Beta on Databricks, accessible via SQL for batch processing or through a REST API for real-time applications and agents. The function is compatible with existing TypeSafe AI API integrations, enabling teams to classify, score, or route data efficiently.

The new function supports practical use cases such as tagging product reviews by issue type and return intent in a single SQL query, or routing user requests to the most suitable AI model based on prompt complexity. For customer support teams, ai_decide can evaluate AI-generated answers against policies to ensure compliance and completeness before deployment.

To demonstrate real-time capability, Databricks built a live demo where ai_decide controls a Snake game by making directional decisions in fractions of a second. This highlights the function’s suitability for applications requiring immediate responses, such as agent actions or application logic, where each decision depends on the current state.