Databricks SDK releases added data quality notification fields, including unhealthy-table alerts

The most concrete shipping change came through the Databricks SDKs. Release v0.191.0 added unhealthy-table notifications for data quality [3]. The Java SDK v0.171.0 added new fields across its data quality API, among others [5], and v0.157.0 added new fields for data quality notifications [7]. All three releases also carried unrelated additions, such as AgentKit memory and session store permission methods and Feature Engineering materialization set methods, so data quality was one part of broader API updates [3][5][7].

Outside Databricks, much of the data quality discussion centered on dbt. Figma's data science team described using ordinary dbt tests as consistency checkers that served as an early warning system for billing bugs, catching more than 100 a month with standard tooling [4]. At dbt Summit 2026, dbt Labs' Alex Noonan argued that grounding is the unsolved part of AI analysts. A dbt Labs benchmark found a model answering through a semantic layer scored 96 to 100% accuracy, against 80 to 91% when the same model wrote raw SQL against a database schema [2].

The community and video content touched the topic more loosely. A Lakeflow demo covered building reliable pipelines, including ingestion from Salesforce with Lakeflow Connect and transformation and orchestration with Lakeflow Designer [8]. Other community posts included one on a transformations API [1], one on why AI data assistants fail without Unity Catalog governance [6], a session on Lakebridge internals and use cases [9], and a list of things the author missed on Databricks [10]. The sources for those posts gave only titles, so this entry does not describe their contents.