AI’s Biggest Opportunity Lies in the 92% of Work It Hasn’t Touched

A new study from MIT’s Center for Collective Intelligence mapped 13,275 commercial artificial intelligence applications against a taxonomy of 20,000 work activities drawn from the U.S. Department of Labor’s occupational database. The finding: 92% of those activities have zero AI coverage. The top 1.6% of tasks — almost entirely content generation and information retrieval — capture more than 60% of all AI market value. The rest of the economy is still waiting.

Where AI Works Today and Why It Stays There

The MIT researchers, led by Patrick J. McGovern Professor of Management Thomas W. Malone, built a 40,000-node ontology of work activities organized into a nine-level hierarchy.

Against that map, they classified every commercial AI tool in the “There’s an AI for That” database alongside 20.8 million robotic systems tracked by the International Federation of Robotics. The resulting picture is sharply skewed.

The single highest-concentration activity, generating images using computers, accounts for 7.18% of all AI software applications. Creating content follows at 3.53%, answering questions at 2.59% and writing content at 1.88%. Together, the top 20 activities, representing just 0.1% of the entire ontology, account for more than 35% of all AI software deployed globally.

The concentration reflects where AI capability matured first. Large language models excel at producing, retrieving and transforming structured text. Those strengths map cleanly onto content workflows, customer service automation and knowledge retrieval, activities where inputs and outputs are well-defined and data is abundant.

As reported by PYMNTS, enterprise AI adoption has accelerated on exactly this foundation, with employees at companies deploying AI tools saving more than an hour per day on tasks like drafting, summarizing and searching.

The 92% That Remains Untouched

The activities outside that narrow band tell a different story. The MIT research identifies zero AI coverage in categories including authorizing decisions, assigning work, collaborating with other actors, complying with directives and analyzing physical objects. These are not marginal functions. They represent the operational core of most enterprises: approvals, resource allocation, coordination across teams, regulatory adherence and the physical environment in which work actually occurs.

The barriers are structural. Activities in these categories depend on fragmented data distributed across systems that were never designed to interoperate, require real-world context that current models cannot reliably interpret and carry accountability requirements that create organizational resistance to automation. Assigning work involves judgment about individual capacity, team dynamics and shifting priorities. Authorizing decisions involves liability. Analyzing a physical environment involves sensor integration, spatial reasoning and situational awareness that text-based models do not natively provide.

As reported by PYMNTS, one of the primary obstacles slowing enterprise AI deployment beyond productivity tools is the challenge of connecting AI systems to the fragmented, inconsistent data infrastructure that underlies core business operations.

The Shift From Assistance to Execution

The MIT framework does not simply describe where artificial intelligence falls short. It functions as a forward map. The researchers categorize the white space into three distinct opportunity types: technical gaps where AI capability does not yet exist, economic gaps where capability exists but has not been productized, and what they call unrecognized opportunities, cases where deployment is both technically and economically feasible but no organization has acted. That third category, the paper argues, may represent the most significant entrepreneurial opening in enterprise AI today.

The path through these gaps runs through agentic systems, workflow automation and deep system integration. PYMNTS has reported that enterprise technology buyers are increasingly evaluating AI not on individual task performance but on whether it can execute multi-step processes with minimal human intervention. That shift moves AI from assistant to operator, from generating a document to managing the workflow that produces, routes, approves and archives it.

The 1.6% of tasks that AI dominates today generated an industry. The 92% it has not reached yet is the larger market. Return-on-investment metrics suggest the technology delivers measurable value. The share of CFO reporting very positive ROI from generative AI jumped from 27% to 85%, according to a PYMNTS Intelligence survey.