Google unveils data cloud purpose built for agentic AI
Features including a Knowledge Catalog and cross-cloud lakehouse are aimed at enabling customers to deploy multi-agent systems and could be a competitive edge for the tech giant.
Google Cloud on Wednesday introduced the Agentic Data Cloud, a new data platform designed to meet the needs of customers as their initiatives evolve from analytics to AI.
The data platforms that enterprises have long deployed to fuel insight-generating initiatives were built to meet the needs of human users such as data scientists, data engineers and data analysts. Increasingly, however, agentic AI applications that reason and act autonomously are becoming the main users of data platforms.
Agents and other AI applications that generate insights and carry out business processes require far more data to be accurate than dashboards, reports and other data products that are designed to be interpreted by humans. In addition, they need data at much higher speeds than analytics tools to act on the most appropriate information.
Built with an AI-native architecture to underpin modern enterprise initiatives, the Agentic Data Cloud features a Knowledge Catalog to provide agentic AI tools with proper context, a cross-cloud data lakehouse that enables AI tools to access data in AWS and Microsoft Azure as well as data in Google Cloud, and a Data Agent Kit.
Kevin Petrie, an analyst at BARC U.S., noted that his firm's research shows that two-thirds of Google users are already putting agents into production. As a result, the Agentic Data Cloud will provide a valuable set of capabilities.
"Google Cloud users are a sophisticated bunch," Petrie said. "Google users need the types of capabilities that Google is announcing this week."
Donald Farmer, founder and principal of TreeHive Strategy, was less enthusiastic about Google's data platform for agentic AI. He noted that although the Knowledge Catalog and cross-cloud lakehouse are valuable for users, their unification in an integrated platform is a necessary change that joins previously disparate capabilities rather than industrywide innovation.
"I see it as a consolidation and rebranding of things they are already using or already announced, now repackaged under an umbrella narrative," Farmer said.
Nevertheless, providing context, eliminating extract, transform and load (ETL) workloads and integrating models, infrastructure and data management are valuable, he continued.
"This announcement is not an industry first in any sense, but there is some differentiation," Farmer said. "The vertical integration of owning models, infrastructure and data systems in one company is true, and it is a real advantage."
The Agentic Data Cloud was unveiled during Google Cloud Next, the hyperscale cloud vendor's annual user conference in Las Vegas.
A data foundation for AI
Google Cloud's current ambition is to drive transformation with its AI-powered technology, according to CEO Thomas Kurian.
Google Cloud users are a sophisticated bunch. Google users need the types of capabilities that Google is announcing this week.
Kevin PetrieAnalyst, BARC U.S.
Toward that end, the tech giant unveiled new features at every level of the AI workflow, from the AI infrastructure layer through research and frontier model capabilities to agents and applications. Included is the Agentic Data Cloud, the data management layer that feeds context to models and agents.
"We're seeing people wanting to delegate tasks and sequences of tasks to agents, and these agents then turn around and be able to use all of GCP as a tool," Kurian said during a virtual press conference on April 20. "All the pieces … are designed to do this [with] our data cloud to feed the agents context from within your systems."
The Agentic Data Cloud represents Google's shift from a platform built for human scale to a platform built for the scale demanded by agentic AI, according to Andi Gutmans, vice president and general manager of Google's data cloud.
The data infrastructures developed before OpenAI's November 2022 launch of ChatGPT sparked surging investments in AI development suffer from structural problems that prevent them from meeting the needs of agentic AI tools, according to Gutmans. Many force users to move their data into isolated lakehouses, fail to provide greater context than a list of data assets and where to find them, and don't provide access to operational systems in real time.
Both the Knowledge Catalog and cross-cloud lakehouse build on existing Google capabilities such as Dataplex and BigLake, while the Data Agent Kit is open source and built for the Agentic Data Cloud. Collectively, they are designed to rearchitect enterprise data infrastructures for AI rather than humans.
"We're making a transition from what I call a system of intelligence to a system of action," Gutmans said. "The first generation of lakehouses and interactions with chatbots was reactive. We're moving toward a world where autonomous agents understand the business. … For agents to reason effectively, we need them to understand the data, how it relates to each other and how it relates to the business."
Google's cross-cloud data lakehouse eliminates data siloes by enabling agentic AI workflows to access data across AWS and Azure to eliminate data egress requirements, including ETLpipeline development and the cost of moving data between platforms.
The Knowledge Catalog adds the context agents require and includes features such as Smart Storage, the LookML Agent and BigQuery Graph -- all of which are in preview. Smart Storage unifies structured and unstructured data, the LookML agent automatically generates semantics that define data and make it easily searched and discovered, and BigQuery Graph enables agents to infer business logic.
In addition, integrations with SaaS applications, operating systems and AI platforms including SAP, ServiceNow and Salesforce enable agents to access live data through the Knowledge Catalog.
Lastly, the Data Agent Kit, also in preview, features Model Context Protocol tools, environment-specific extensions and prebuilt plug-ins that enable developers to bring Google's agentic AI development tools into platforms such as VS Code and Claude Code. In addition, the Data Agent Kit features prebuilt agents for data engineering, data science and data observability.
Most features are either in preview or expected to be generally available soon, according to Google.
While needed to meet modern enterprise needs and logically constructed do so, there are more capabilities Google's new data platform for agentic AI could include, according to Farmer.
"Everything announced sounds coherent, but evaluation and observability get short-changed, at least in the announcement," he said.
Petrie, meanwhile, suggested that the capabilities that are included in the Agentic Data Cloud address some of the problems enterprises face when developing agentic AI tools and could help Google attract new customers.
"The Agentic Data Cloud takes logical steps to meet strong customer demand in areas such as cross-cloud integration, cross-stack agent orchestration and federated cataloging," he said. "I think this release will help Google gain incremental market share in the AI space against Azure and AWS."
In particular, cross-cloud data integration stands out as a valuable capability, Petrie continued.
"The cross-cloud integration … is a big deal," he said. "Nearly half of AI adopters have data in hybrid or multi-cloud environments, so they really need federated access. I'll be interested to see how this feature is adopted when it goes GA."
Beyond the Agentic Data Cloud, new capabilities introduced during Google Cloud Next include AI infrastructure improvements such as a new generation of tensor processing units to handle the scale and latency requirements of AI tools, new agent security capabilities, and the Gemini Enterprise Agent Platform for agent development, orchestration and governance.
Looking ahead
Beyond observability and evaluation capabilities, Farmer advised that Google -- which is the developer of the Agent2Agent Protocol for managing agents in production -- add multi-agent coordination capabilities in the Agentic Data Cloud.
"They don't say … what happens when you have a thousand agents and they need to coordinate, share resources, deal with conflicts or understand what each other is doing," he said. "Managing and ensuring that a large number of agents work together is a complex problem that vertical integration alone can’t solve."
Gutmans noted that Google already plans to address some of the concerns raised by Farmer with its roadmap for the Agentic Data Cloud.
"In the Knowledge Catalog, we believe we're going to need to have evaluation be part of that," he said. "That's not part of our launch this week, but it's definitely an area that's super critical. As the Knowledge Catalog keeps getting enriched, we need to make sure that none of the existing workflows take a negative turn."
Agent orchestration, meanwhile, is included in AI workflow capabilities rather than the Agentic Data Cloud, Gutmans continued.
In addition, Google's product development plans include connecting AlloyDB to its cross-cloud lakehouse, enabling Spanner to work outside of Google Cloud, and adding AI capabilities to the interconnected nature of its AI infrastructure, models and data platform, according to Gutmans.
"You're going to see us truly make the Agentic Data Cloud a unified data cloud so customers don't have to assemble the pieces and stitch them together," Gutmans said.
Eric Avidon is a senior news writer for Informa TechTarget and a journalist with more than three decades of experience. He covers analytics and data management.
