Particle.news

Databricks Recasts Lakehouse as Foundation for Agent-Driven Enterprise Workflows

The company rolled out a coordinated set of products to run governed, real-time AI agents on a single lakehouse platform.

Overview

  • Databricks announced the package of updates at its Data + AI Summit on June 16–17, centering its strategy on a single lakehouse layer that serves both live transactions and analytics for agentic applications.
  • The new LTAP (Lake Transactional/Analytical Processing) architecture stores transactional and analytic data in one open lake layer so agents can read fresh data without copying it to separate systems.
  • Lakebase now offers Postgres-compatible branching and real-time mirroring that let agents fork full databases quickly for safe experiments while mirrored columnar copies power fast analytics.
  • Lakehouse//RT, powered by the Reyden engine, aims to deliver millisecond query latency directly on Delta and Iceberg tables so high-concurrency agent workloads can serve and act on live data.
  • Databricks paired an expanded agent stack (Agent Bricks, Genie/Genie Code and scheduled autonomous tasks) with Unity Catalog AI Gateway for runtime governance and added Microsoft OneLake and NVIDIA Vera/GPU integrations to support scale, security, and cost control.