Google Cloud adds cross-cloud data access and business context for AI agents

Google Cloud is giving AI agents new ways to reach business data and interpret what it means. Its November 4, 2026 Agentic Data Cloud update combines cross-cloud access, automated knowledge discovery and developer tools. Data Agent Kit and BigQuery Graph are now generally available, while several capabilities remain in preview.

The platform first launched earlier this year at Google Cloud Next, with a design that lets agents access data where it lives rather than requiring a centralized warehouse. Google’s rationale is that autonomous agents will become the main consumers of data. Sirish Chandrasekaran, its vice president of product management for data analytics, says that shift calls for infrastructure built around fast access, less data movement and automated governance.

Queries cross cloud boundaries

The new borderless Lakehouse builds on Apache Iceberg, an open-source table format. Chandrasekaran says it can query remote data in Amazon Web Services and Microsoft Azure over secure private links without egress fees. It caches retrieved data blocks inside Google Cloud so later queries can run faster. The mechanism therefore includes a local cache, rather than leaving every retrieved block exclusively in the original cloud.

A separate AlloyDB integration lets users query external Iceberg tables using standard PostgreSQL queries, without copying the underlying data. That connects current transactional records with historical lakehouse data through the database’s existing query language.

Business meaning becomes agent context

Access is one part of the update; interpreting an organization’s information is another. Knowledge Catalog, Google’s map of business meaning for agents, is getting self-learning enrichment. The preview continuously scans unstructured material in Google Workspace, Jira, SharePoint and GitHub. It turns that operational knowledge into context sets in Open Knowledge Format, giving agents organizational information to draw on when answering questions or carrying out work.

Chandrasekaran says early adopters saw agents become four times more accurate on average, while token consumption fell by up to 75%. Those are Google’s early-adopter performance claims. The company also says Knowledge Catalog is being enabled by default in the Gemini application starting November 4.

Within Gemini, the catalog is meant to resolve business-metric definitions and access controls before a prompt becomes SQL, the language used to query databases. Companies can also ground agents in semantic models—definitions of business measures and relationships—from BigQuery, Snowflake, SAP, Looker and dbt.

Whether an agent is calculating net revenue for finance or customer churn for operations, it applies the exact same governed formulas your business teams rely on.

Sirish Chandrasekaran, Google vice president of product management for data analytics

From stated goals to running pipelines

Google is also adding autonomous machine-learning workflows to Gemini. Data engineers can describe a desired business outcome, and Gemini will generate PySpark code, run notebooks and fix pipeline errors as it works toward the goal. For business users, Gemini can construct and save operational-reporting queries using BigQuery and Knowledge Catalog. Google says those saved reports can then run on demand without further token costs.

• Generally available: Data Agent Kit brings Model Context Protocol tooling, which connects agents to external tools and context, into development environments including VS Code, Claude Code and Antigravity. Developers can describe their intent and have agents create, refine and test data pipelines.
• Generally available: BigQuery Graph brings petabyte-scale graph analytics into relational database tables without a separate extract, transform and load pipeline, according to Google.
• In preview: TimesFM 3.0 adds forecasting, while TabFM addresses predictive analytics.

Database capacity is receiving attention too. AlloyDB Express is entering broader preview, with provisioning in under two seconds, Google says. PostgreSQL for agents in AlloyDB provides an isolated workload environment that the company says can handle three million queries per second. These infrastructure changes target high-concurrency use: many agents querying a database at the same time.