Oracle adds business definitions and simulations to its AI lakehouse

Oracle wants AI agents to understand company data, not just retrieve it. On October 26, it announced Autonomous AI Lakehouse updates combining shared business definitions, broader data operations and what-if simulations. The promise is business-aware decision support across connected systems; Oracle has not specified a release date for the new capabilities.

The company’s approach is to query data where it already lives, rather than require copies in a separate AI environment. Oracle says the platform connects to Oracle and non-Oracle sources, including Snowflake and Databricks. Its AI Data Catalog can connect to AWS Glue, Databricks Unity Catalog and Snowflake Horizon Catalog, bringing discovery, business context and governance together across those sources.

Giving business terms a shared meaning

The Business Semantic Layer supplies common definitions for users, applications and agents. Customers can annotate data, import existing Oracle Database SQL annotations and generate AI-enriched descriptions of tables and columns. Oracle says that shared context helps tools interpret data consistently across connected systems, rather than leaving each agent or analytics tool to work from disconnected descriptions.

For more complex calculations, AI-suggested Oracle Analytic Views package metrics, hierarchies and business calculations into reusable models. They also contain the logic for joining and aggregating data, reducing the query logic users must write. Oracle says support for Apache Ossie lets customers import compatible semantic models instead of recreating existing definitions.

The catalog also narrows what an agent needs to inspect. Oracle describes a search process that moves from catalogs and schemas down to relevant tables and columns, using business meaning and relationships. The intended benefit is more focused context, with less metadata for an agent to process.

In Oracle-published analyst commentary, theCUBE co-founder and chief analyst Dave Vellante said the combination could ground agent workflows in relevant business metrics and relationships. He also identified a prerequisite:

Realizing that value requires accurate business definitions and clear controls over execution.

Dave Vellante, co-founder and chief analyst, theCUBE, in commentary published by Oracle

Beyond reading data

Expanded Apache Iceberg support goes beyond querying and analysis. Oracle says users and agents can create, populate, update and delete Iceberg tables through Oracle SQL, its database query language. Support for Iceberg V2 and V3 broadens compatibility, while customers can retain their supported catalogs, object storage and compatible processing engines.

Access controls sit in the AI Data Catalog. Oracle describes role-based permissions covering distributed Oracle databases, selected non-Oracle sources and Iceberg tables. Customers can authorize access to an individual Iceberg table, for example, rather than grant access to an entire object-storage bucket.

Data Studio brings this work into an AI-assisted workspace accessible through an Oracle Cloud Infrastructure account. Oracle describes three practical routes through it:

• Discover and analyze data: search cataloged assets, connect to more than 100 data sources, and work through natural language, SQL or Python.
• Build data pipelines: describe integration goals to the AI Assistant, then create and refine flows that cleanse, join, enrich and publish data.
• Manage operations: use the assistant to monitor and debug jobs, inspect database and cloud infrastructure state, and consult product documentation.

Modeling a decision before committing

The new business simulation engine uses Oracle Essbase for planning, forecasting and what-if analysis. Users can model changes in revenue, demand, pricing, capacity, cost or risk, with calculations across business hierarchies. Results can be written back to Oracle AI Database and Iceberg tables. Business users can ask natural-language questions, generate simulation reports or analyze models in Excel through Oracle Smart View.

Oracle also claims faster processing. It reports a threefold Data Lake Accelerator improvement in its TPC-DS database benchmark on a 100 TB dataset, alongside an increase in maximum cluster size from 32 to 100 nodes. Lake Cache stores frequently accessed external data on Exadata flash; in Oracle’s Iceberg TPC-DS test, query time fell from 96 seconds to 17 seconds.