For decades, data professionals have struggled to manage operational and analytical databases separately, facing latency and performance degradation. With the advent of AI agents, the problem has become structural: a system that reasons continuously and acts on live data cannot tolerate a pipeline between itself and the information it needs. At the Data + AI Summit this week, Databricks announced two products aimed at collapsing that infrastructure: Lakehouse//RT and LTAP (Lake Transactional/Analytical Processing).
Lakehouse//RT: Millisecond Latency Without a Dedicated Serving Tier
Lakehouse//RT eliminates the need for a separate real-time serving tier by enabling sub-100ms latency queries directly on governed Delta and Iceberg tables. The Reyden compute engine, built for high-concurrency, low-latency serving, handles up to 12,000 queries per second, with response times as low as 10ms on smaller datasets and up to 16x better performance than traditional stacks. Every query runs within the Unity Catalog governance framework, without data copies or ingestion pipelines.
Sponsored Protocol
LTAP: Storage-Layer Unification
LTAP is Databricks' answer to the failure of HTAP (Hybrid Transactional/Analytical Processing) approaches. Instead of converging query engines, LTAP unifies data at the storage layer via the Lakebase architecture, a serverless PostgreSQL service. Transactional data is written directly in Delta or Iceberg format, eliminating the ETL pipelines that have connected operational and analytical systems for decades. As co-founder Reynold Xin explained, the goal is to use the best engine for each workload while sharing a single copy of data on underlying storage. The key engineering challenge is latency: object storage has response times in seconds, too slow for OLTP. Lakebase uses a caching layer between Postgres compute instances and object storage, where idle CPU converts rows to columns, achieving over 10x compression and reducing network costs.
Sponsored Protocol
Analysts see the agentic framing and open format approach as differentiators. Stephanie Walter from HyperFRAME Research notes that agents need live operational data, historical context, governance, retrieval, and write-back in the same workflow. Mike Leone of Moor Insights and Strategy adds that letting transactional writes land in open formats is a less common move, making a credible case for retiring specialized systems. However, the challenge remains to prove that both engines truly share a single copy without hidden conversion steps.
For enterprises, the question is no longer which tool to choose for each workload, but whether maintaining separate tools is still defensible. Data from VB Pulse Q1 2026 shows hybrid retrieval intent tripled, while standalone vector database adoption declined. Even in virtualization, companies seek unified solutions, as shown by HPE's offer of free virtualization software to attract dissatisfied VMware customers. Similarly, Anthropic's feud with the Trump administration highlights how market dynamics can influence AI adoption. In this context, Databricks' proposal to eliminate copies and synchronization between operational and analytical systems represents a significant step forward for agentic workloads. For a historical overview, consult Databricks on Wikipedia.
Sponsored Protocol
