MIT survey finds AI agents access only 45% of corporate data, leaders far ahead

Data gap that decides agents

A survey of 300 data and technology executives conducted by MIT Technology Review Insights in partnership with Google Cloud reveals the primary bottleneck for AI agents: access to information. On average, agents have access to only 45% of an organization’s data. In firms classified as “data laggards,” the figure falls to 30% or lower, while “data leaders” expose agents to more than 70% of the information. The gap is not theoretical; it determines whether an agent can make an informed decision or merely guess.

Trust built on infrastructure, not promises

Only about half of the surveyed organizations say they trust the decisions made by their agents. Among the leaders, the figure is 100%. The report notes that agents operating on partial, outdated, or context-free data are likely to err. Moving from answering questions to executing actions requires real-time access to operational systems, supply-chain data, point-of-sale information, human-resources records, and legacy systems, even those refreshed only a few years ago, which are not built for such use.

Bottlenecks disappear among leaders

Two-thirds of the laggards (66%) report that outdated data systems limit their ability to scale agent deployments, and 68% say the same constraints hinder rapid decision-making. Among the leaders, who have largely resolved legacy constraints, only 8% cite each of these obstacles. The conclusion is clear: the difference between a stalled pilot and a broad rollout is not the model but the infrastructure feeding it.

Two years to prepare, or guaranteed failure

All respondents (100%) plan to use AI agents within two years, and 69% anticipate a wide rollout. Gartner forecasts that agents will augment or automate 50% of business decisions by 2027, adding urgency. Without removing data bottlenecks, the speed and efficiency promised by the technology will not materialize. Organizations that understand this are already investing in three areas: improving access to structured and unstructured information, strengthening data and AI governance with business context, and automating data management itself.