The missing layer between AI hype and real business value

Every company seems to be using AI now. Very few are getting transformative results from it. That gap — between widespread adoption and genuinely useful deployment — is the focus of a new MIT Technology Review Insights survey, and its findings suggest the problem is less about the models themselves than about what those models actually know.

Widespread use, modest payoff

According to the survey, every organization polled uses AI to some degree. Yet only 16% say they are extracting transformative value from their deployments. The largest group, 40%, reports moderate gains confined to specific tasks or teams, while 21% report meaningful gains and 23% say they are seeing minimal returns or none at all.

The pattern shows up in how companies are actually deploying the technology. About 38% of respondents say they are mostly still piloting use cases, and 35% say they are mainly using AI to assist employees through co-pilots or chat assistants. Just 12% have AI agents in production, and only 3% say AI does most of the work in key processes.

The capability gap

Prukalpa Sankar, co-founder of Atlan, puts the problem bluntly in the report: there is a wide gap between what models can do and how useful they are inside organizations. In her view, the raw capability is already there — what is missing is the connective tissue that would let those models operate intelligently within a specific company.

That connective tissue, Sankar argues, is context: a structured understanding of how a business actually works — its systems of record, its data, its definitions, its rules. A context layer, as described in the report, sits between an organization's business systems and its AI platforms, translating enterprise knowledge into governed, machine-usable form so that agents and models can act on it reliably. Without it, she says, deploying a powerful model inside a company is like hiring a brilliant newcomer, blindfolding them, and expecting them to do useful work on day one.

Why this matters for AI budgets

The survey's message cuts against the usual explanation for disappointing AI returns — that the technology simply is not ready yet. If the bottleneck is context rather than capability, then the fix is architectural and organizational: companies need to invest in the layer that connects their data and systems to their AI tools, with governance built in, before expecting agents to deliver.

That also reframes the next wave of AI spending. Pilots and assistants can show value in narrow tasks without much context. But the bigger prize — agents handling real processes end to end — depends on whether the AI understands the business it is working in. Until that layer exists, the report suggests, most companies will keep getting modest returns from very capable tools.

Written by the BidAsk Newsroom. Based on reporting from the sources below.