The Gap Between Capability and Usefulness
Walk through almost any large organization today and you will find a crowded landscape of artificial intelligence at work. Customer service platforms answer questions around the clock, conversational assistants draft replies, and automation tools shuffle data between systems without complaint. Each piece is impressive on its own. Yet the results often feel uneven, with outputs that are confidently wrong, responses that ignore company policy, or actions that make sense in isolation but collide with how the business actually runs.
A recent MIT Technology Review Insights poll sharpens the picture. Every surveyed organization is using AI to some degree, but fewer than one in five report transformative value. Most describe moderate gains in specific tasks or teams, while a meaningful slice sees minimal returns. Only a small share have AI agents running in production, and just three percent say AI handles most of the work in key processes. The technology is clearly present. The payoff is lagging.
Why Context Is the Missing Layer
Prukalpa Sankar, co-founder of Atlan, argues that the bottleneck is not raw model intelligence. Frontier systems already perform impressively on a wide range of reasoning and language tasks. The shortfall lies in how well those systems understand the specific business they are supposed to serve. Which customer segment matters most this quarter? Which approvals require a second signature? Which data definitions differ between finance and operations? Without answers to questions like these, even a brilliant model behaves like a new hire who has been handed no onboarding materials.
This is where the idea of a context layer comes in. Sitting between systems of record and the AI tools that draw on them, a context layer translates scattered enterprise knowledge into governed, machine-readable information. Done well, it does more than feed facts to a model. It captures the rules, relationships, and history that make a company distinctive, and it allows that knowledge to accumulate over time rather than being rebuilt for every new tool. Sankar’s blunt analogy, that deploying AI without context is like hiring the smartest intern in the world and blindfolding them, captures the stakes neatly.
What Comes Next for Enterprise Buyers
The next phase of enterprise AI will likely be decided less by which model is newest and more by which organizations can make their knowledge legible to machines. Expect growing attention to data governance, metadata management, and shared business definitions, areas that once seemed unglamorous but now sit at the center of AI strategy. Companies that treat context as infrastructure, rather than an afterthought bolted onto a pilot, are more likely to move from scattered experiments to agents that reliably complete real work. For technology leaders weighing where to spend next, the question is shifting from “Which AI should we buy?” to “How well does our AI understand us?” That shift in priorities will shape vendor selection, budgets, and the tools that eventually reach everyday business users, and it is a trend worth watching closely as adoption deepens across industries.
