Exclusive: Google Cloud accelerates shift to agentic data
Google Cloud is repositioning its data platform strategy around what it calls an "agentic" future, where artificial intelligence agents - rather than humans - become the primary users of enterprise data systems. The shift reflects changing expectations around how organisations build, manage and interact with data as AI capabilities mature.
At the centre is a new "agentic data cloud" concept, alongside product launches focused on real-time context, automation and cross-cloud access. The strategy has evolved over several years, following the progression from analytics to generative AI and now multi-agent systems.
Agent shift
"From our perspective, especially in the Data Cloud world, we're talking about the agentic Data Cloud," said Yasmeen Ahmad, Product & GTM Executive - Data & AI Cloud at Google Cloud, Google.
"The goal has been to figure out what the future data platform needs to be to support this agentic era, this agentic future, and fundamentally, there are some big differences in how humans operate with data platforms versus agents," said Ahmad.
Data systems are moving beyond human-centric design. Autonomous agents, capable of executing tasks, require platforms built for machine interaction, speed and precision.
"We believe the future data platform is going to be used by agents. It's not being used by humans," said Ahmad. "We're very quickly accelerating to a world where we've got agent as the central persona."
This shift changes how data must be structured and delivered. The model centres on three transitions: from raw data to semantic knowledge, from reactive intelligence to proactive action, and from human-scale to agent-scale operations.
Context layer
A central component is the "knowledge catalogue", a universal context engine designed to aggregate and enrich information from across the enterprise. It brings together data from multiple platforms, including SAP, Salesforce, Workday, ServiceNow and Palantir, alongside internal documents and systems.
"What we discovered over the last three years was context is absolutely critical for enterprise use cases," said Ahmad. "Models are trained on the world of data, but they know nothing about a business, their metrics, their intuition, their unwritten rule book."
The system extracts context from both structured and unstructured sources, such as policies, spreadsheets and documents. It then applies search and retrieval techniques to deliver only the most relevant information to each agent.
"Search and retrieval is critical, because if you serve up too much context, agents suffer from context rot and their performance degrades," said Ahmad.
Access controls are built in, ensuring agents retrieve only data permitted under organisational policies.
"Just like your data historically has access control and policy, the same applies as we bring the knowledge catalogue online," said Ahmad.
Real-time data
The approach prioritises dynamic, real-time data rather than static, pre-trained models. As business data changes, the context layer updates continuously.
"This context is dynamically living, like your business is constantly evolving, changing, morphing," said Ahmad. "So the knowledge catalogue is a continuously evolving context engine."
Consistency is critical in enterprise use cases, where outputs must be reliable.
"When it comes to enterprise data and questions, you can't have a different answer every five minutes," said Ahmad. "It needs to be the right answer every time."
Agent-driven workflows illustrate this model. Multiple agents handle specific tasks, each receiving narrowly defined context, before combining results into a final output.
"Each agent needs just the right context for the operation that it's proposing," said Ahmad.
Tooling push
To support development, Google Cloud introduced Data Agent Kit, combining tools, skills, plugins and extensions that allow agents to interact directly with its data ecosystem.
"The Data Agent Kit has tools, skills, plugins and extensions," said Ahmad. "These essentially allow an agentic environment to understand Google's agentic Data Cloud natively."
Automation is also being applied to previously manual processes, including vector embedding management, reducing the need to build and maintain custom pipelines.
"Rather than having to manually hand-code pipelines and create vector embeddings, now it's just being managed by the data platform," said Ahmad.
This reflects a broader shift towards abstraction, where infrastructure complexity is handled by the platform.
Performance gains
Core data engines are being updated to handle increased workloads from AI agents. Improvements to BigQuery include a "fluid scaling" capability designed to boost efficiency and reduce costs.
"With BigQuery, we announced fluid scaling delivering a 34% faster efficiency of querying," said Ahmad.
Enhancements to Apache Spark workloads also aim to improve price-performance compared with alternatives such as Databricks.
"Agents can do 10 to 20 query requests in the time a human would send one," said Ahmad. "We want the engines to be more performant and cost-effective to scale to that volume."
Cross-cloud access
To address multi-cloud environments, Google Cloud introduced Cross-Cloud Lakehouse that enables data access without migration between platforms such as AWS, Microsoft Azure and Snowflake.
"You don't move the data, we connect it," said Ahmad.
The approach relies on cross-cloud interconnect and open standards such as Apache Iceberg to enable interoperability.
"In sub-second timeframes, you can connect large volumes of data between these clouds," said Ahmad.
By removing the need for large-scale migrations, the model aims to shorten implementation timelines.
Broader access
The shift towards agentic systems is not limited to large enterprises. Advances in AI models, including Gemini, are lowering barriers to entry.
"What I find exciting about the agentic era is regardless if you're a startup with five people or a large multinational, everybody has access to these really powerful models," said Ahmad.