Glonce

Epoch AI finds 20 percent of US workers now delegate tasks to AI

researchThe Glonce editorsThe Decoder🔥60Glonce rating60Impact55Novelty45Relevance87Surprise42Players52Credibility76
Epoch AI finds 20 percent of US workers now delegate tasks to AI

Epoch AI and polling firm Ipsos surveyed 1,106 employed US adults between July 10 and 19, 2026, and found that 20 percent now delegate at least one work task to AI that a coworker or an outside contractor used to handle. Respondents were asked about ten common work tasks drawn from US Department of Labor data and chosen to reflect typical knowledge work; workers reported using AI on all ten, though how much varied widely by task. Where AI is used, workers generally accept its output with little to no editing rather than treating it as a rough draft that needs heavy rework.

Adoption is highest in computer systems and software development, where 57 percent of workers who do that task use AI for it, the highest of the ten tasks studied. Data analysis follows at 46 percent, then reading work documents at 39 percent; record-keeping sits at the bottom of the ten at 25 percent. But using AI for a task rarely means AI does the whole job: full or near-full completion by AI reaches 10 percent only in software development and stays below 7 percent for every other task in the survey.

Task substitution, meaning AI has taken over work a person used to do rather than just assisting with it, is most common in data analysis at 7.1 percent, followed by reading work documents at 5.7 percent and record-keeping at 5.3 percent. Epoch AI stresses that this kind of substitution does not automatically mean the affected workers are being displaced entirely; the researchers frame the overall pattern as tasks moving between humans and AI rather than whole jobs being automated away.

How much of a task AI handles tracks with how much time workers say they save. When AI provides only partial help, respondents report saving time on 37 percent of those tasks; when AI does most or all of the work, that figure rises to 53 percent. The survey cannot say whether heavier AI use actually makes people faster or whether workers simply reach for AI more when they already want to save time. AI does not reliably speed things up either: about one in six AI-assisted tasks now takes longer than before, which the researchers suggest may happen because interacting with AI itself takes time, or because workers use the time it frees up to do the task more thoroughly or to a higher standard.

On how workers treat what AI produces, 66 percent of AI output gets used unchanged or with only minor tweaks, though just 6 percent is used with no changes at all; on the other end, 5 percent gets heavily reworked or mostly rewritten. The researchers caution that low editing effort is not a direct measure of AI output quality, and they found no consistent link between how much time workers reported saving and how much editing they did to the output.

Epoch AI sums up its findings by describing AI as a versatile but usually not self-sufficient workplace tool, something that most often trims or reshapes a task rather than replacing the person doing it. The results rest on self-reported data; neither actual time savings nor the quality of AI output were measured objectively, and the ten tasks studied were chosen based on national employment data rather than on how likely each one is to be affected by AI.

The finding sits alongside two earlier surveys that point in the same direction without measuring the same thing. 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. A more recent Anthropic survey of roughly 9,700 Claude users found that about half of respondents believed AI could already handle 50 percent or more of their own work, but that sample was drawn from users of one specific AI product rather than a representative slice of the workforce, so it is not directly comparable to the Epoch AI and Ipsos numbers.

Key facts

  • Epoch AI and Ipsos surveyed 1,106 employed US adults between July 10 and 19, 2026, and found that 20 percent now delegate at least one work task to AI that a coworker or outside contractor used to handle.
  • AI adoption is highest in software development, at 57 percent of workers who do that task, and data analysis, at 46 percent, but full or near-full task completion by AI reaches at most 10 percent and stays below 7 percent for every other task studied.
  • Task substitution, AI fully taking over work a person used to do, is highest in data analysis at 7.1 percent, followed by reading work documents at 5.7 percent and record-keeping at 5.3 percent.
  • Workers report saving time on 37 percent of tasks where AI helps only partially, rising to 53 percent when AI does most or all of the work; about one in six AI-assisted tasks takes longer than before.
  • 66 percent of AI output is used unchanged or with only minor edits, and the researchers found no consistent link between reported time savings and how much editing workers did.

Why it matters

Most surveys about AI at work ask whether people use it at all. This one asks what changes once they do, and puts a number on it: 20 percent of employed Americans now hand off at least one task to AI that a colleague or an outside contractor used to do. The task-level breakdown is what makes the finding useful rather than just another adoption statistic. Software development and data analysis lead in adoption, and even there, AI fully or near-fully completing a task stays at 10 percent or below, so the picture is real but uneven and still mostly partial. That is a more concrete read on where AI actually sits inside white-collar work today than a single top-line usage percentage can give.

Who it affects

Most directly, the 1,106 employed US adults surveyed, and by extension the wider population of US knowledge workers whose jobs touch the ten tasks studied: software development, data analysis, reading work documents, record-keeping, and others the source does not name individually. Software developers see the highest AI adoption of any task in the survey, and data analysts see both high adoption and the highest rate of AI fully taking over their work. It also affects the coworkers and outside contractors whose work is being redirected to AI, and employers weighing whether AI adoption is actually translating into time saved, since the survey finds that link is real but inconsistent.

How to use it

There is no product here, only a data point for anyone deciding how to use AI at work or how to read claims about it. The clearest practical signal is that heavier AI use correlates with more reported time savings: 37 percent of partially AI-assisted tasks save time versus 53 percent when AI does most or all of the work, though the survey cannot say whether that link is causal. It also shows why light review is currently the norm: two-thirds of AI output needs no more than minor edits before use. But AI use is not a guaranteed shortcut; about one in six AI-assisted tasks takes longer than the task took before, so applying AI to a task does not by itself guarantee time saved.

How solid is it

The findings come from a survey described as representative, conducted by Epoch AI with polling firm Ipsos across 1,106 employed US adults between July 10 and 19, 2026, using ten tasks drawn from US Department of Labor data. All figures are self-reported: neither actual time savings nor AI output quality were measured independently, and the source gives no detail on sampling weights or margin of error. The pattern lines up directionally with two earlier, differently designed surveys, a Gallup poll from August 2025 that found 45 percent of US workers use AI on the job, and an Anthropic survey of roughly 9,700 Claude users where about half thought AI could already handle 50 percent or more of their work, but neither is a repeat wave of the Epoch AI and Ipsos survey, and the source draws no time trend from its own data alone.

Risks and caveats

Every number here is self-reported, not independently verified, so the survey shows what workers believe happened, not measured output quality or measured time savings. The correlation between heavier AI use and reported time savings could mean AI genuinely saves more time the more it is used, or it could mean workers turn to AI more when they already want to save time; the source says this cannot be determined from the data alone. Low editing effort is not evidence of high output quality either, and the researchers found no consistent relationship between how much workers edited AI output and how much time they said they saved. The ten tasks studied were chosen to match national employment data, not because they were expected to be the tasks most affected by AI, so the pattern may not extend to work the survey did not cover. Epoch AI's own framing, that this is task redistribution rather than job elimination, is the researchers' interpretation of the data, not a separate measurement.

“versatile but usually not self-sufficient workplace tool.”

— Epoch AI