Databricks introduces LTAP (Lake Transactional/Analytical Processing), a new architecture that eliminates the distinction between transactional (OLTP) and analytical (OLAP) databases. Storage is consolidated into a single open format, without ETL pipelines. With Lakebase as its foundation, Databricks is preparing enterprise data infrastructure for an era in which AI agents will become the primary users.
Databricks announced this during the Data + AI Summit in San Francisco. The company notes that for forty years, there has been a clear divide between OLTP and OLAP systems. Transactional databases keep production systems running, while analytics systems process historical data for reporting. Companies bridge that gap with ETL pipelines, replicas, and specialized data engineers. But that model doesn’t hold up when thousands of AI agents simultaneously switch between the two systems to read live transactional data, analyze historical context, and take action.
Databricks now aims to permanently bridge that gap with LTAP. The approach unifies OLTP and OLAP storage into a single layer.
No pipelines, no replicas
Previous attempts to bring OLTP and OLAP together ran into trouble. HTAP systems resulted in high costs, performance trade-offs, and vendor lock-in. “Zero ETL” turned out to be a hidden CDC mechanism in practice: still two copies of data, the same issues regarding freshness and isolation.
HTAP takes a different approach. Data is stored once in an open format on object storage, with standard Postgres semantics. Dedicated compute engines serve both transactional and analytics workloads, ensuring performance isn’t sacrificed for either. Databricks builds on the Lakehouse architecture it previously introduced as an open and unified foundation for data and AI.
Lakebase as the foundation for AI agents
Lakebase extends LTAP to mission-critical workloads. The Postgres-compatible, serverless database adds native vector search and full-text search, eliminating the need for a separate vector database. Real-time event ingestion occurs via Lakeflow Zerobus. Object storage enables Git-style branching, allowing agents to spin up their own database environments for safe experimentation without affecting the production system.
Agents read operational data the moment a transaction is recorded, complete with full historical context. No replicas, no ETL, no risk to production.
According to Databricks, AI agents will become the primary users of the enterprise data stack. The existing infrastructure, built for people running reports, is not designed for this. LTAP and Lakebase are the answer to this anticipated shift.