Sep 5 — Oct 5, 2026
Over the past month, Databricks focused heavily on operationalizing AI agents, highlighted by the rollout of the Unity AI Gateway CLI, Model Context Protocol (MCP) support for Genie, and new low-latency decision functions in SQL.
1.Genie One establishes an enterprise ontology layer and MCP integration
Databricks pushed Genie One into production workflows with the general availability of the Genie One Model Context Protocol (MCP) server and Unity Catalog Pages. These additions standardize business definitions and data relationships into a governed semantic layer that external agent harnesses and internal business users can query directly.
Sources
- From Data to Dialogue: How S&P Global Energy Made Its Structured Data Estate Conversational with Databricks Genie Agents and MCPNews · databricks-blog · Sep 25
- Genie One MCP: Give any AI Agent the Right Business ContextNews · databricks-blog · Sep 22
- The Genie One MCP is now Generally AvailableNews · databricks-blog · Sep 22
- Unity Catalog Pages: a governed home for your business knowledge in Genie OntologyNews · databricks-blog · Sep 22
- The Ontological Definition of Databricks Genie Ontology via OntoRankVideo · Advancing Analytics · Sep 21
2.Unity AI Gateway expands agent governance with the ug CLI and MCP tooling
Following the general availability of Unity AI Gateway, Databricks introduced the Unity Gateway CLI (ug) to centrally configure, route, and budget coding agents and LLM workloads. Accompanying SDK releases and Terraform provider updates added automated credential management and securable types for external MCP servers and agent skills.
Sources
- How Enterprises Govern AI Agents Across Multiple ModelsVideo · Databricks · Oct 1
- v0.157.0Release · databricks/databricks-sdk-java · Oct 1
- The Unity Gateway CliVideo · Databricks · Sep 26
- Deploy and manage coding agents at scale with the Unity Gateway CLINews · databricks-blog · Sep 24
- v1.133.0Release · databricks/terraform-provider-databricks · Sep 21
3.In-engine AI classification debuts with the launch of ai_decide()
The introduction of the ai_decide() SQL function brings fast, structured categorization and scoring directly to data warehouses without requiring generative LLM invocations. The capability is supported by SDK updates and practitioner workflows using specialized judge models to evaluate pipeline outputs at lower latency and cost.
Sources
- Let AI Decide: SQL Decision Making with Databricks ai_decide()Video · databricks MVP Hubert Dudek · Oct 3
- v0.158.0Release · databricks/databricks-sdk-java · Oct 1
- Introducing ai_decide: make fast decisions on your governed dataNews · databricks-blog · Sep 30
- v0.144.0Release · databricks/databricks-sdk-py · Sep 30
- Running open-Jev in SQL on DatabricksNews · databricks-blog · Sep 24
4.Lakebase Postgres adds native search, branch restores, and LTAP patterns
Databricks broadened its transactional offering with the general availability of Lakebase Search for combined semantic and keyword indexing within Postgres. Platform tooling also added metadata-driven branch restores for near-instant multi-terabyte recovery, CLI lifecycle controls for ephemeral databases, and zero-ETL integrations with analytical lakehouse tables.
Sources
- Lakebase Postgres branch-based restores for fast recovery at scaleNews · databricks-blog · Oct 1
- A practical guide to cost optimization with Lakebase PostgresNews · databricks-blog · Sep 30
- v1.19.0Release · databricks/cli · Sep 30
- Lakebase Search: State-of-the-art full text and vector search for PostgresNews · databricks-blog · Sep 28
- Big Data LDN 2026 - Lakebase & LTAP | Simon Whiteley & Holly SmithVideo · Advancing Analytics · Sep 24