Scaling AI agents with trustworthy data: New findings from MIT Technology Review Insights

As the agentic era redefines what’s possible for organizations of all sizes, the way we’re working is shifting dramatically.

To capitalize on this movement, there needs to be a parallel shift in how to build data systems. They should be optimized to activate all enterprise data, driving the highest quality outcomes, while doing so at the lowest possible cost.

To understand how businesses are navigating this transition, Google Cloud partnered with MIT Technology Review Insights to publish a new report, Scaling AI agents with trustworthy data. The report shares findings from a global survey of 300 data and technology executives and features in-depth interviews with leaders from HCA Healthcare, Shopify, and Deutsche Telekom.

The research makes one thing very clear: Successfully scaling AI agents depends on the strength and readiness of your data foundation.

While nearly all surveyed organizations already use or plan to use AI agents, most deployments remain small-scale, with only 10% of organizations using them widely across their business today. However, this landscape will look completely different very soon. Within two years, roughly two out of every three organizations, or 69%, plan to deploy AI agents widely.

This suggests a significant leap is on the horizon for many organizations, as they go from early pilots to widespread, enterprise-grade operations.

The report illuminates the gaps that many organizations still face in getting their data ready for AI, as well as making sure those AI systems can reliably access the information they need to achieve real-time, trusted action across the enterprise. The report also identifies areas where organizations are deploying AI agents and finds that companies with AI-ready data systems are achieving the strongest results, offering a blueprint for others to follow.

Continue reading to learn more, or check out the full report for greater detail and guidance.

Data leaders vs. data laggards: Trust starts at the data layer

To move past early pilots and scale agents widely, organizations must address their underlying data systems first.

Many of us are impressed by the growing capabilities of AI agents. To make good decisions and take effective action, agentic systems need a data foundation that is multimodal, context-aware, and instantly available. Legacy data systems struggle to meet these demands, ultimately compromising AI trustworthiness.

In fact, more than half of respondents (55%) say legacy data systems are preventing them from scaling agentic AI across their enterprise.

Surveyed executives point to four main factors that constrain their existing data platforms from supporting AI agents:

• Entrenched silos: When data sits in disconnected systems, agents cannot see the whole picture of the business.
• Difficulty accessing and managing unstructured data: Teams struggle to safely unlock valuable dark data hiding in PDFs, emails, videos, and call logs.
• Insufficient access to real-time data: Old batch-processing architectures only show agents what happened yesterday, making real-time action challenging.
• A lack of business context and semantics: True context goes beyond basic metadata, allowing agents to understand exactly what the data means and how different assets relate to each other. Without this deep understanding, data cannot be highly relevant to specific use cases.

While these legacy constraints hold many companies back, a distinct group is breaking through.

The research reveals that the organizations having the most success are “data leaders,” which the report defines as those organizations who give their AI systems access to more than 70% of their enterprise data. “Data laggards,” on the other hand, only provide access to 30% or less of their data for AI.

It’s clear that for most companies, data access is still a huge roadblock. Among the organizations surveyed, on average they only grant access to 45% of their enterprise data for AI purposes. This impacts how much teams trust their AI: Around half of the surveyed executives overall (51%), and just one-fifth of those who qualified as data laggards (22%), trust the accuracy and relevance of their AI agents’ decisions.

By contrast, this trust gap effectively vanishes for data leaders. A little more than one third (36%) of these organizations report that their agents are “consistently accurate,” while the remaining 64% say their AI is “mostly accurate” when providing outputs and decisions. None report “moderately accurate” or “inaccurate” results, compared to three-quarters of laggards who do.

By opening up access to most of their data, these leading organizations see successful outcomes with agents across many business functions. They also find it easier to scale agents to make accurate decisions at speed. This clearly demonstrates how building a reliable AI system starts with unlocking trust at the data layer.

Evolving for the agentic era

So, how do organizations actually evolve and get ready for the agentic era? Surveyed organizations identify the following top three data initiatives as key to scaling AI agents over the next 12 months:

• Improve access to all types of data: Teams connect disparate systems to activate both structured and unstructured data right where it lives.
• Improve data and AI model governance with business context: They enrich models with true business context, empowering agents to act with accuracy across the enterprise.
• Replace batch processing with streaming and event-driven pipelines: Organizations speed up their data architectures so agents can react and make decisions in real time.

Ultimately, bolting AI onto an old architecture is slow, expensive, and frustrating for teams. Because a patchwork approach lacks deep integration, it can lead to spiraling costs and quickly become a financial liability.

Many in the industry still build these kinds of "Systems of Intelligence," but the goal is no longer just to know; the goal is to act. To do this safely and effectively, organizations need a System of Action.

At Google Cloud, we believe the solution is an Agentic Data Cloud. This approach evolves your data foundation from a static, fragmented repository into a dynamic, unified knowledge platform. To power this shift, your architecture must meet three requirements: It must be AI-native to manage complex reasoning in real time; borderless to activate data across any cloud and format; and fundamentally trusted to deliver accuracy, governance, and deep business context.

The team would like to thank Googler Ashish Chopra for his contributions to this post.