No longer call it Autonomous Data Warehouse, but Autonomous AI Lakehouse.
Oracle is making this brand shift in light of several functional evolutions. Among them, native Iceberg support, giving rise to the lakehouse concept. The addition of an agentic framework likewise justifies the AI dimension.
The Iceberg integration is initially certified for AWS Glue, Snowflake Polaris, Databricks Unity and Apache Gravitino. The SQL syntax of Autonomous AI Database is evolving in parallel, to enable queries such as SELECT * FROM owner.table@catalog.
From Select AI RAG, to Select AI Agent
The agent component is named Select AI Agent. It continues the trajectory of Select AI, launched in late 2023 under the banner of text-to-SQL.
Since then, Select AI has been equipped, among other things, with a RAG (Retriever-Augmented Generator) module designed to enrich natural-language queries. More recently, Oracle released a Python port.
So it now opens up to agentic AI, backed by a framework ReAct*. It reuses the RAG component, paired with MCP compatibility and the ability to leverage external tools via REST (web search with the OpenAI API, in particular). Several guardrails are in place, including LLM-as-a-judge to evaluate outputs and the possibility to define “SQL profiles” linked to user-defined rules.
Table Hyperlink, the New Name for Pre-authenticated URLs
The rebranding of Autonomous Data Warehouse into Autonomous AI Lakehouse runs alongside another change: the feature previously known as PAR URLs (Pre-Authenticated Request URLs) becomes Table Hyperlink.
The pre-authenticated URL system enables temporary REST-based access to tables or views within Autonomous Database. These URLs, generated by executing PLSQL code, can have an expiration date and/or a maximum number of uses. They can also be manually invalidated. Since their launch in early 2024, they have been enhanced in several respects. For data producers, the ability to extend the validity window of URLs via a few API calls; and a “selective sharing” system allowing access to subsets of datasets over the Internet while keeping the rest inside a Virtual Cloud Network (VCN). For data consumers, the web UI has improved, with color-coding to identify trends and anomalies.
The Table Hyperlink branding is meant to better reflect the feature’s objective (connecting tables to workflows). Oracle promises to integrate, in the future, default association variables, ensure consistency for paginated URLs… and, most importantly, enable the management of multiple tables with a single link.
In the realm of external data processing, Oracle has integrated into its database a flash memory cache system (in Exadata). Supporting Parquet, ORC, AvRO files and Iceberg tables, it is for now manual (the user must designate the tables or parts of tables to cache). There is talk of automating the process based on usage analysis.
AI Data Platform, In the Footsteps of MySQL HeatWave Lakehouse
The lakehouse dimension is not immediately apparent in the branding of AI Data Platform, yet it lies at its core. The offering, which has just reached general availability, is an evolution of an existing product: MySQL HeatWave Lakehouse. It builds upon Autonomous AI Database, Oracle Analytics Cloud (with the option to connect to third-party BI tools), as well as object storage and OCI’s generative AI services (access to models from Meta, Cohere, xAI, etc.). The compute layer runs on Apache Spark, paired with NVIDIA GPUs. In this sense, the whole differs from Autonomous AI Lakehouse, which is more analytics-oriented.
Autonomous Data Warehouse and AI Data Platform underpin another offering, not entirely new but also born from the branding shift. It is Fusion Data Intelligence, formerly Fusion Analytics Warehouse. It enables Oracle’s analytics tools to work in concert with Fusion Cloud applications, delivering a complete pipeline, a data warehouse, a semantic model, and ready-to-use content (metrics, workbooks, visualizations).
* In broad terms, the ReAct approach weaves together chain-of-thought generation and action planning, invoking human feedback when necessary.
