AI tools can respond quickly and perform sophisticated tasks, yet still produce actions that do not fit a company’s needs. The article points to a missing ingredient: business context that helps AI systems understand the organization they work in.
AI use is widespread, but results vary
An MIT Technology Review Insights poll found that every surveyed organization was using AI to some degree. Still, only 16% said they were getting transformative value from their deployments.
More respondents reported narrower results. 40% saw moderate gains in specific tasks or teams, and 21% reported meaningful gains. Another 23% said returns were minimal or absent.
Most organizations were still testing AI or using it to assist employees. 38% said they were piloting use cases, while 35% used tools such as co-pilots or chat assistants to support work. In contrast, 12% had AI agents in production, and 3% said AI was doing most of the work in key processes.
Why capable tools can miss the mark
Businesses may have several AI platforms, systems, and tools operating at once, including customer platforms, conversational assistants, and workflow automation. Each tool can be advanced and responsive, but its output may still be confused, incorrect, or disconnected from how the business operates.
Prukalpa Sankar, co-founder of Atlan, describes a gap between what models can do and how useful they are inside organizations. In the article, she argues that broad model capability alone does not ensure practical value in a company.
Without relevant information about the business, an AI system may not know how its task fits company processes or what organizational knowledge should guide its response. That can limit the usefulness of AI investments, even when the underlying tools are powerful.
A context layer connects business knowledge to AI
The proposed solution is an architectural context layer between business systems and AI platforms. It takes enterprise intelligence from sources such as systems of record and data, then makes that information governed and usable by AI models or agents.
In practical terms, this layer is intended to give AI tools a clearer picture of the company while they work. It can help align their outputs with business knowledge and support knowledge that compounds as it is made available to AI.
Sankar compares an AI system without this context to “the smartest intern in the world” being blindfolded and placed inside an organization. The analogy captures the central argument: intelligence is less useful when the system cannot see the information that would help it act appropriately.
From pilots to more useful deployments
The poll figures suggest that many companies are still building experience with AI through pilots and employee assistance tools. Production agents and AI doing most of the work in key processes remain less common among respondents.
A context layer is presented as infrastructure that could help close that gap by connecting existing business knowledge to AI deployments. The article does not claim that context alone guarantees returns; rather, it frames context as a missing part of making AI more accurate, cost-effective, and useful within an enterprise.
For organizations evaluating AI, the core question is therefore not only which model or platform to adopt. It is also whether the systems can draw on governed company knowledge, and whether that knowledge is available in a form AI can use.