Databricks expanded workspace AI functions with SDK support, new services, and the ai_decide method.

Databricks expanded programmatic control over AI functions throughout September 2026. Early in the month, platform releases introduced SDK support for AI Functions [7] and formally established AI Functions, Domains, and Sandboxes as workspace services [6]. Toward the end of the month, release v0.144.0 expanded this workspace service further by adding the ai_decide method, giving users a direct mechanism for decision-oriented operations alongside environment variable updates across jobs and tasks [1].

Practical patterns for applying these capabilities to production data were also highlighted. Guidance demonstrated how teams can use ai_query within SQL queries and Lakeflow jobs to evaluate custom model endpoints, such as SemIf-OpenJev running on serverless GPUs, returning structured outputs and probability scores at scale [2]. Alongside this, technical sessions detailed how developers can configure GenieCode with workspace instructions, custom skills, and MCP servers to enforce data standards during SQL pipeline generation [4].

Within the community, practitioners worked through how to implement these capabilities responsibly. Teams debated designs for AI-driven data testing systems within automated DevOps promotion paths to satisfy management mandates without introducing unreliable checks [5]. At the same time, informal discussions touched on changing industry vocabulary and the shift toward super intelligence branding [3].