Databricks released the Omnigent agent harness and introduced Lakebase Postgres for agentic workloads.

Databricks introduced Omnigent, an open-source meta-harness designed to orchestrate, sandbox, and govern multiple AI agents within unified sessions [23], [37]. To prevent data exfiltration across agent toolchains, Omnigent introduced contextual security policies that track private data access, untrusted content exposure, and outbound transfer, cutting off egress before unauthorized transfers occur [21]. Databricks contributed Omnigent to the Open Secure AI Alliance [33] and integrated it alongside Model Context Protocol servers and Databricks Apps to standardize deployment patterns [27], [32].

Database architecture also shifted to accommodate the high-frequency reads and writes produced by autonomous agents. Databricks detailed Lakebase Postgres, which decoupled compute from storage by pairing object storage with a write-ahead log to address agentic input-output bottlenecks [2], [4]. In parallel, Electric joined Databricks to incorporate WebAssembly-based Postgres into lightweight agent sandboxes, allowing local agent databases to sync back to central Lakebase storage [20].

Platform governance expanded to manage agent sprawl across tools and semantic definitions. Unity Catalog extended centralized discovery, access control, and auditing to agents and MCP servers [10], while Pages in Unity Catalog added curated business definitions for Genie agents to cite [6], [14]. In production use cases, Databricks reported its internal AI SRE agent managed over 2,000 incident investigations daily [7], and Document Intelligence added an agentic harness to execute multistage extraction on complex documents [15]. Meanwhile, competitive evaluations at the Grounded Reasoning Cup underscored that domain-tailored agents with verification outpaced generic foundation models on specialized reasoning tasks [16].