Oracle Pitches Autonomous AI Lakehouse as Open Semantic Layer for Real-Time Agent Queries

Oracle Corp. is revamping its enterprise data stack around autonomous artificial intelligence systems, positioning its Autonomous AI Lakehouse as an open semantic layer that federates queries across heterogeneous cloud environments and competing data platforms without requiring data duplication.

The architectural shift reflects an operational change in enterprise computing: as autonomous software agents join human analysts to execute automated workflows, traditional batch pipelines and scheduled data replications struggle to provide the low-latency information those models need to make decisions.

Speaking at the Oracle AI World conference, Nipun Agarwal, senior vice president of software engineering for Oracle Database, said lakehouses are transitioning from historical reporting tools into operational decision engines for software agents, in an interview first reported by SiliconANGLE.

Central to Oracle's design is a federated catalog architecture that mounts third-party metadata repositories, including Databricks Inc.’s Unity Catalog, as well as relational engines such as PostgreSQL and MySQL. Rather than extracting and loading records into a centralized Oracle repository, the platform executes queries against the source systems where the data resides.

At an event in October, Oracle expanded its support for the Apache Iceberg open table format and introduced a business semantic layer. Under this framework, automated descriptions generated for tables and columns apply across both native Oracle stores and external repositories, providing language models with standardized business definitions across disparate schemas.

To address the latency penalties typically associated with federated queries, Oracle is deploying its Lake Cache technology, which stages frequently accessed Apache Iceberg tables directly into Exadata flash memory. Agarwal noted that real-time responsiveness has become more critical because software agents generate unpredictable, on-demand query patterns that outpace traditional batch-refresh intervals.

Oracle insists the strategy is not intended to displace established lakehouse deployments. Instead, the company expects large organizations to continue operating multi-vendor architectures spanning several lakehouses, with its semantic layer functioning as a connective fabric across existing investments.

The real test for Oracle's federated model will come as enterprise customers expand autonomous agent deployments into production environments, where distributed query performance and metadata consistency across non-Oracle catalogs will determine whether in-place processing can match the reliability of dedicated analytical stores.