Genie One MCP reached general availability to provide governed context to AI agents.

Databricks made the Genie One Model Context Protocol (MCP) server generally available, allowing external agents to query enterprise data with governed context from Genie Ontology [14, 15]. The integration exposed approved metrics, data definitions, and access permissions across structured and unstructured data [15, 25]. Concurrently, Lakebase Search reached general availability on AWS and Azure, adding hybrid keyword and vector search to Postgres without standalone ETL pipelines to support agent memory and operational retrieval [2, 20].

Infrastructure for agent state and execution received significant attention. Databricks detailed Lake Transactional and Analytical Processing (LTAP) architectures to allow agents to write and query operational state alongside analytical tables [8, 38]. Tooling for database branching, including the Consort framework, introduced isolated environments for agent tasks and safe regression testing [19, 33]. In addition, Databricks published the Big Book of AgentOps to document production architectures for observability, cost management, and tool policies [42, 43].

Cost control and governance remained central concerns for production fleets. Teams used Unity AI Gateway to track model spending in system tables and diagnose runaway tool calling loops [11, 36, 44]. Case studies demonstrated rigorous evaluation patterns in production: Zepto detailed customer support agents running behind strict MLflow quality gates [34], and Concurrence governed clinical agents operating at trillion token scales [12]. Practitioners also evaluated safe write paths and investigated methods to build context modeling directly inside warehouses [6, 13, 39].