BigQuery and Oracle bring AI agents to analytics

In October, Google BigQuery made new AI functions and conversational agent improvements production-ready, and Oracle announced a semantic layer so its agents understand the business. What it means for companies.

Asking your data a question in plain language ("which customers stopped buying this quarter?") and getting a reliable answer is the promise of AI agents for analytics. In October 2026, Google BigQuery and Oracle took concrete steps in that direction.

Google BigQuery: AI functions ready for production

October 13: the AI.AGG function, which uses AI to summarize and group unstructured information such as comments, emails or descriptions, became generally available.
October 14: conversational analytics agents can now query data sources across several projects and be published in Gemini Enterprise.
October 15: agent answers, with text, charts, tables and citations, can be exported to PDF or CSV.
October 21: AI.CHUNK_DOCUMENT, which splits documents into chunks for AI search (RAG), became available for production.
October 26: conversations with the agents can be saved to a BigQuery table, which helps audit what was asked and what was answered.

Oracle: a semantic layer so AI understands the business

Oracle announced a business semantic layer for its Autonomous AI Lakehouse: reusable models that define dimensions, measures, hierarchies, calculations and key performance indicators (KPIs), so that natural language queries always follow the same business rules. The company said the capability will arrive within the next 12 months.

Oracle made the announcement on October 26 at its AI World event in Las Vegas, together with broader support for data in the Apache Iceberg format.

What it means for your company

AI agents make data easier to query, but they do not fix messy data: if "sales" means something different in each department, the agent will answer with the same confusion. Before connecting an agent, it is worth:

Defining your key indicators in writing, along with how they are calculated.
Consolidating data into a single source and reviewing its quality.
Giving each user access only to what they need and keeping a log of queries.
Starting with one concrete business question and checking that the agent's answer matches your reports.

October 31 update: Oracle's official announcement

We checked this article against Oracle's press release. We removed the case of a pharmaceutical client cited by a media outlet: it was told by an outside consultant and referred to a different platform, not Oracle's.

How we help you

At Pronetsys we design data analytics and artificial intelligence projects: we organize your sources, define your indicators and build dashboards and agents that answer with reliable figures. If you want to know where to start, book a consulting session.