One in five US workers now delegates tasks to AI

Between July 10 and 19, 2026, Epoch AI and polling firm Ipsos surveyed 1,106 employed US adults about their use of AI on the job. The survey drew ten common tasks from US Department of Labor data, chosen to reflect typical knowledge work. The headline result: 20 percent of respondents say they now hand off at least one task to AI that a human colleague or contractor used to perform.

AI use varies sharply by task type. In computer systems and software development, 57 percent of workers who do that task use AI. Data analysis reaches 46 percent, reading work documents 39 percent, and record-keeping trails at 25 percent. However, AI adoption does not mean full automation. Workers mostly describe AI as partial support. Full or near-full task completion by AI occurs only in software development, at 10 percent, and drops below 7 percent for all other tasks. When respondents were asked about task substitution—where AI has taken over work that people used to do—data analysis leads at 7.1 percent, followed by reading documents at 5.7 percent and record-keeping at 5.3 percent. Epoch AI emphasizes that this task substitution does not automatically imply worker displacement.

Time savings show a correlation with the intensity of AI work. When AI handles only part of a task, respondents report saving time in 37 percent of cases. When AI does most or all of the work, that figure rises to 53 percent. Yet AI does not always accelerate work: about one in six AI-assisted tasks now takes longer than before. The researchers suggest this may occur because interacting with AI itself consumes time or because workers use the freed-up capacity to perform tasks more thoroughly or at higher quality.

Workers typically accept AI output with light or no revision. 66 percent of AI output is used unchanged or with only minor edits, and just 6 percent is used without any changes at all. At the other end, 5 percent of output gets heavily reworked or mostly rewritten. The researchers note that low editing effort is not a direct proxy for output quality and found no consistent link between reported time savings and editing effort. The survey results rest on self-reported data; neither actual time savings nor AI output quality were objectively measured. The tasks were selected based on national employment data rather than on their likelihood of being affected by AI. Earlier surveys point in a similar direction: a Gallup survey from August 2025 found that 45 percent of US workers use AI on the job, though only 10 percent use it daily, and a recent Anthropic survey of roughly 9,700 Claude users showed that about half believed AI could already handle 50 percent or more of their work—though that sample consisted of users of a specific AI product, not a representative population cross-section.