AI adoption at work is broad but shallow
A new economic study argues that generative AI's presence in the workplace is wide-ranging but not particularly deep, with most workers in AI-exposed roles not actually using the tools regularly. Researchers found that while genAI touches around 80% of occupations and over 40% of tasks, fewer than half of workers in most of these occupations have adopted it, suggesting earlier estimates from AI vendors overstated how embedded the technology has become in everyday work.
The paper, "What Work Does Generative AI Do?", was produced by economists from the Federal Reserve Bank of St. Louis, Vanderbilt University and Harvard, using US Real-Time Population Survey data rather than chat logs from AI firms like OpenAI, Microsoft and Anthropic, which the authors say inflate exposure figures through overly generic task classifications. Adoption is highest in management, finance, business and computing roles, and lowest in personal service and manual jobs; only around one in six occupations exceed 70% adoption, and no task category surpasses that threshold. As of May 2026, 45% of US working-age adults used generative AI for work purposes and 62% used it overall, with the researchers noting that prior experience with AI in one area strongly predicts adoption elsewhere.
- GenAI reaches most jobs, but usage within them stays shallow
- Vendor exposure estimates overstate real-world AI adoption, study finds
- 45% of US workers use genAI for work; 62% use it overall
New here? Start with this
Generative artificial intelligence, or genAI, refers to tools that can produce text, images or other content in response to instructions. At work, people may use them to draft documents, analyse information or help with computer tasks.
How much these tools affect a job depends on whether its tasks could be done with AI and whether workers actually use it. Researchers have estimated that many occupations include tasks suited to genAI, but estimates based on AI company usage data may not show how often workers use the tools in practice.
Use varies across types of work: it is more common in management, finance and computing, and less common in personal service and manual jobs. The distinction matters because a technology can be relevant to many jobs without becoming a regular part of most workers’ daily routines.
Both sides, in good faith
The strongest fair case each way — we don't pick a winner.
The case for
Broad exposure to generative AI may still matter even when regular use is limited: workers can apply it selectively to particular tasks, and familiarity gained in one area may encourage wider adoption. The study also suggests use is already substantial in some knowledge-based roles, so the technology could be changing work unevenly rather than being absent from everyday jobs.
The case against
The findings support caution about claims that generative AI is already deeply embedded across the workforce: most workers in AI-exposed occupations do not use it regularly, and adoption remains low in many roles. Estimates based on broad task classifications or company chat data may overstate practical workplace use, so sweeping claims about AI’s current impact could get ahead of the evidence.