Scaling AI agents with trustworthy data
MIT Technology Review MIT Technology Review Insights ● Covered by 4 sources
AI agents are spreading fast, but most companies still don’t give them enough data to work with. That’s why the gap between hype and real ROI is turning into a data problem.
Based on reporting by MIT Technology Review, MIT Technology Review Insights — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
The promise of agentic AI is no longer the hard part. Business and technology leaders already see where this is headed: software that doesn’t just answer questions, but takes action inside the company. The snag is much less glamorous. Many firms are finding that the return on that promise depends on a data setup they don’t actually have.
MIT Technology Review’s report, based on a survey of 300 data and technology executives, says the biggest blockers are legacy systems and weak data foundations. Agents need more than a clean dashboard or a few approved feeds. They need access to structured and unstructured data across the enterprise, plus enough business context to make sense of it. They also need direct reach into operational systems such as supply chain, point-of-sale, and human resources data. Older systems, even ones refreshed only a few years ago, are still getting in the way.
The numbers in the report make the problem look stubborn. Across the surveyed organizations, AI agents have access to an average of 45% of company data. In the data laggards, that falls to 30% or less. The data leaders are doing the opposite: they open up more than 70% of their data to agents, and they’re seeing better results because of it. Trust tracks the same way. Only around half of respondents trust their agents’ decisions today, while 100% of the data leaders do.
And the speed issue is just as stark. Two-thirds of data laggards say legacy systems limit scaling, and 68% say those systems keep agents from making decisions fast enough. Among the leaders, only 8% report either problem. That is a pretty neat way to separate the companies that are merely experimenting from the ones that are actually ready to let agents do work.
The pressure is only going up. Within two years, every respondent expects to be using agentic AI, and 69% expect to use it widely. The report’s takeaway is blunt: if companies don’t make their data estates agent-ready, the technology won’t deliver the speed and efficiency everyone is banking on. The priority list is equally plain: better access to structured and unstructured data, stronger governance with business context, and more automation in data management.
My take — AI-written commentary, not fact-checked reporting
This is the part of AI nobody gets to skip: the pipes matter more than the demo. A company can buy all the agents it likes, but if the data is trapped, dirty, or context-free, the machine is basically doing improv in a locked room. The winners here won’t be the loudest AI adopters; they’ll be the ones doing the boring governance work everyone else keeps postponing.
Read more about this at: MIT Technology Review