Google Cloud updates Agentic Data Cloud to support large-scale autonomous AI agents

Artificial intelligence is no longer just about conversational assistants that tell people what to do. These days, it’s all about work automation, and Google Cloud is making sure it’s right at the forefront of that trend with its revamped Agentic Data Cloud platform, which received a host of new updates today.

The Agentic Data Cloud was launched earlier this year at Google Cloud Next. It’s meant to be the architectural “system of action” for AI agents, replacing the brittle extract, transform and load-based data pipelines, or ETL, that feed legacy software applications. The key advantage of Agentic Data Cloud is that it meets customers where their data is, so rather than forcing them to squeeze all of their information in a large, centralized data warehouse, it enables AI agents to access it wherever it lives.

In a blog post, Sirish Chandrasekaran, Google’s vice president of product management for data analytics, outlined the company’s belief that autonomous agents rather than humans will soon become the primary consumers of data. As such, it makes sense to build a purpose-built infrastructure stack specifically for them, prioritizing low latency, zero data movement and automated governance. “That is why we built the Agentic Data Cloud — a system of action designed to activate all your data by breaking down walled gardens, deliver high accuracy and trust, and operate proactively at low cost,” Chandrasekaran said.

The flagship new feature in Agentic Data Cloud is borderless Lakehouse, a special version of Google’s “cross-border Lakehouse” product that also debuted at Cloud Next. Chandrasekaran said borderless Lakehouse is built upon the open-source Apache Iceberg table format and designed to eliminate multicloud friction with new caching capabilities for Amazon Web Services and Microsoft Azure. This allows companies to query remote data held in those cloud platforms over a secure private link, without having to pay egress fees.

Chandrasekaran detailed how it works, saying that it automatically caches retrieved data blocks within Google Cloud to accelerate subsequent queries. There’s also a new AlloyDB integration that allows users to query external Iceberg tables with standard PostgreSQL queries without having to copy data. In this way, it bridges real-time transactional data with historical lakehouse records.

More grounded agents

One of the big problems with autonomous agents is that they can be prone to hallucinations without proper guardrails in place, which is why Google has overhauled another capability within Agentic Data Cloud. The Knowledge Catalog, which is a kind of map of business meaning that’s meant to inform AI agents of the peculiarities of the organization they work for, is getting a new self-learning enrichment capability in preview.

According to Chandrasekaran, it works by continuously scanning unstructured data in Google Workspace, Jira, SharePoint and GitHub to translate operational knowledge into Open Knowledge Format context sets. These can be used to ground AI agents in organizational knowledge, and the results from early adopters shows that it can help make the average agent four-times more accurate, while reducing token consumption by up to 75%.

Knowledge Catalog is seen as so important to Google’s agentic future that it’s also being enabled by default in the Gemini application starting today. Because it’s so tightly integrated, it can resolve issues around business metric definitions and access controls before it translates user’s prompts into Structured Query Language queries. Companies can also choose to ground agents in semantic models for specific data platforms, including Google BigQuery, Snowflake, SAP, Looker and dbt.

“Agents are only as accurate as the business context they’re grounded in,” Chandrasekaran said. “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.”

Gemini is also getting new autonomous machine learning capabilities that can help to streamline everyday work for both technical teams and business workers. For instance, data engineers can now tell Gemini a specific business outcome they want to achieve in natural language, and Gemini will generate the PySpark code, run the notebooks and fix any data pipeline errors as it works toward that goal. Meanwhile, business users will be able to generate reliable, real-time operational reports in Gemini simply by asking for them.

“Gemini uses operational reporting skills integrated with BigQuery and the Knowledge Catalog to construct and save the query,” Chandrasekaran explained. “Once saved, teams can run reports on demand without incurring token costs, guaranteeing consistent and verified answers every time.”

Accelerated agentic infrastructure

Google also has a few gifts for software developers, starting with the new Data Agent Kit, which is generally available from today, bringing Model Context Protocol tooling into integrated development environments such as VS Code, Claude Code and Antigravity. This means that developers can now just define their business intent in natural language and have agents automatically create, fine-tune and test the associated data pipelines.

Developers can also take advantage of infrastructure updates that are designed to facilitate “high-concurrency agent interactions,” so that thousands of AI agents can query the same database simultaneously without swamping it and causing problems. These include the broader preview of AlloyDB Express, a highly scalable database that supports sub-two-second provisioning for rapid agent prototyping. There’s also PostgreSQL for agents in AlloyDB, which provides an isolated workload environment that can handle three million queries per second.

Finally, Google said BigQuery Graph is now generally available, bringing petabyte-scale graph analytics into relational database tables with zero ETL. It’s joined by previews of the TimesFM 3.0 forecasting model and TabFM predictive analytics model.

Google Cloud’s Yasmeen Ahmad, managing director of product management, data and AI cloud, sat down with SiliconANGLE founder Dave Vellante at Google Cloud Next, during an exclusive broadcast on theCUBE, SiliconANGLE Media’s livestreaming studio, at Google Cloud Next, where she discussed how Google is reimagining the data stack for AI agents:

Image: SiliconANGLE/Gemini