Databricks Unifies Data for AI Agents

Databricks announced two new products, Lakehouse//RT and LTAP, at the Data + AI Summit, targeting the long-standing challenge of unifying operational and analytical data without performance degradation. These innovations aim to collapse existing data infrastructure, providing millisecond query latency directly on governed Delta and Iceberg tables with Lakehouse//RT. LTAP stores Postgres-native transactional data in Delta and Iceberg format from the point of write, removing decades-old ETL pipelines.

Reynold Xin, co-founder of Databricks, highlighted that a simpler data stack is crucial for AI agents to move faster. LTAP represents Databricks' approach to HTAP (Hybrid Transactional/Analytical Processing) by unifying data at the storage layer rather than through engine convergence. This method leverages the Lakebase architecture, ensuring a single copy of data is accessible to both transactional and analytical workloads.

Lakehouse//RT addresses the need for a dedicated real-time serving tier by enabling direct querying of live lakehouse data with sub-100ms latency. This capability, powered by the Reyden compute engine, operates within Unity Catalog's governance framework, eliminating data copies and complex ingestion pipelines. Analysts note that the agentic AI framing and open format approach are key differentiators, though Databricks must prove the latency and reliability expected by CIOs.