Databricks announced Lakehouse//RT and LTAP at the Data + AI Summit in San Francisco, two products designed to collapse the infrastructure between operational and analytical data. For decades, data professionals have struggled with managing separate databases that introduce latency and performance degradation. Agents made this problem structural: a system that reasons continuously and acts on live data cannot tolerate a pipeline between itself and the information it needs to act on.
The goal is to eliminate the 'middleman' in your data stack. Databricks is moving away from the traditional approach of using different tools for different workloads. By unifying storage at the lakehouse level, companies can deploy more reliable AI agents faster, leading to more efficient workflows and reduced technical debt.
Lakehouse//RT delivers millisecond query latency directly on governed Delta and Iceberg tables. This removes the need for a dedicated real-time serving tier, which typically creates data copies and split governance. Key capabilities include:
- Reyden compute engine: Specifically built for high-concurrency, low-latency serving by querying tables directly.
- Extreme performance: It delivers sub-100ms latency at 12,000 queries per second.
- Improved speed: It offers response times as low as 10ms on smaller datasets, which is 16x better performance than existing dedicated serving stacks.
LTAP simplifies how you ingest transactional data into your lakehouse. Short for Lake Transactional/Analytical Processing, it stores Postgres-native transactional data in Delta and Iceberg format from the point of write. This removes the ETL pipelines that have connected operational and analytical systems for years. Instead of converting data after it lands, LTAP ensures transactional data lands directly in open formats, sharing the same copy that analytical workloads read.
Storage-layer unification is the 'holy grail' for agentic AI. Databricks co-founder Reynold Xin argued that as users 'vibe code' more applications, the underlying infrastructure needs to get out of the way.
The agents really prefer a much simpler stack, because they can move way faster,
Xin said in a briefing with VentureBeat.
The shift toward unified storage is already accelerating. Data suggests a move away from standalone vector databases toward hybrid retrieval. According to VB Pulse Q1 2026, hybrid retrieval intent tripled from 10.3% to 33.3% in a single quarter. For your team, this means the question is no longer which best-of-breed tool to run for each job—it is whether running separate tools is still defensible. If your agents are finding inconsistencies across governance boundaries, these products suggest you can potentially retire separate real-time serving tiers and complex ETL pipelines today. Read more: Snowflake signs $6B AWS chip deal — why your cloud costs will drop this fall.
