Databricks added five Predictive AI Functions, including ai_forecast, ai_top_drivers and ai_detect_anomalies, in October 2026.

The main announcement of the month was the Predictive AI Functions release on October 14. Databricks introduced five functions: an upgraded ai_forecast, plus the new ai_top_drivers, ai_detect_anomalies, ai_predict_value and ai_predict_class. They were aimed at SQL analysts and business users, who could generate predictions, detect anomalies and find the drivers behind a metric inside Genie and existing workflows [4].

Document processing was the second theme. Two videos covered Intelligent Document Processing with the built-in AI functions. The first, on October 12, explained how AI parse, AI classify and AI extract pull information from PDFs and images and bring it into the lakehouse [5]. The second, on October 16, was a demo of an end-to-end flow. Documents were uploaded to a volume, parsed, classified by type, and then used for data extraction and automated workflows [3]. A community article on October 6 applied the same approach to property and casualty claim documents, turning them into structured data [6].

The community also spent time on ai_decide. One article on October 4 described it as a judge that returns a decision rather than generated text [7]. Another, on October 1, used it as a low-cost router placed in front of ai_query, choosing between GLM 5.3 Flash and GLM 5.3 for each request [8]. A third article on October 22 showed how to generate realistic synthetic test datasets using AI functions in plain SQL [2].

AI functions also came up outside the core SQL topics. A video on October 26 described the Lakebase vector and text extensions for hybrid search in Lakebase Postgres, aimed at agent retrieval without an external database [1].