AI is no longer only a tool workers use alongside familiar office software. For a meaningful share of employed Americans, it has become a substitute for another person on at least some tasks.
A representative survey by Epoch AI and Ipsos found that 20 percent of employed Americans now delegate at least one work task to AI that was previously handled by coworkers or outside contractors. The finding is narrow but important: it does not say that whole jobs are being automated away, but it does show that parts of knowledge work are being reassigned.
What the survey measured
The survey interviewed 1,106 employed US adults between July 10 and 19, 2026. Respondents were asked about ten common work tasks, selected from US Department of Labor data to reflect typical knowledge work.
Workers reported using AI across all ten tasks, although the level of use differed sharply from one type of work to another. The results show a workplace where AI adoption is broad, but uneven, and where human judgment still remains part of most workflows.
The central result is that one in five working adults in the US says AI now handles at least one task that used to be assigned to a coworker or outside contractor. That makes AI delegation a practical workplace behavior, not just a theoretical productivity trend.
Where AI is used most
AI use is highest in computer systems and software development. Among workers who perform that task, 57 percent use AI. Data analysis follows at 46 percent, while reading work documents reaches 39 percent.
Record-keeping is at the lower end of the survey results, with 25 percent of affected workers using AI for that task. Even there, the number suggests that AI has entered routine administrative work, though less aggressively than in technical or analytical tasks.
The survey also makes an important distinction between using AI and handing over an entire task. In most cases, workers describe AI as partial support rather than a complete replacement. Full or near-full task completion by AI reaches 10 percent only in software development and remains below 7 percent for every other task.
That distinction matters for how employers and workers interpret the data. AI is being used in many places, but the survey does not show a broad pattern of machines fully completing entire categories of work without human involvement.
Task substitution is visible, but limited
The clearest shift from human to machine work appears in data analysis. In that area, 7.1 percent of respondents say AI has taken over tasks that people used to handle. Reading work documents follows at 5.7 percent, and record-keeping comes next at 5.3 percent.
Those numbers are smaller than the headline adoption rates because they focus on substitution, not general use. A worker may use AI to assist with data analysis without treating it as a replacement for a coworker or contractor. The survey separates those ideas, which helps clarify what is actually changing.
Epoch AI stresses that task substitution does not automatically mean workers are being displaced entirely. The data points to a more granular shift: some parts of work are moving from people to AI systems, while the broader job may still remain with the employee.
AI saves time for some workers, not all
The survey found a relationship between how much of a task AI handles and how often workers report saving time. When AI handles only part of a task, respondents report time savings on 37 percent of those tasks. When AI does most or all of the work, the figure rises to 53 percent.
That does not prove that heavier AI use causes workers to become faster. The survey data alone cannot determine whether AI creates the time savings or whether workers turn to AI more heavily when they are already trying to save time.
The results also show that AI can introduce friction. About one in six AI-assisted tasks now takes longer than before. The researchers suggest two possible explanations: interacting with AI may take time, or workers may use the extra capacity to complete tasks more thoroughly or at higher quality.
This mixed picture is important. AI delegation is not simply a universal shortcut. It may help with some tasks, slow others down, and change what workers spend their attention on.
Most AI output is lightly edited
Another notable finding is how workers treat the output they receive. According to the survey, 66 percent of AI output is used unchanged or with only minor tweaks. Only 6 percent is used with no changes at all, while 5 percent is heavily reworked or mostly rewritten.
That suggests many workers are comfortable accepting AI-generated material after limited review. At the same time, the researchers note that low editing effort is not the same thing as proven output quality. The survey did not objectively measure whether the AI output was accurate, useful, or better than the human alternative.
The researchers also found no consistent link between reported time savings and the amount of editing workers performed. In other words, lightly edited output did not automatically mean more saved time, and heavier editing did not tell the whole story either.
Epoch AI describes AI as a "versatile but usually not self-sufficient workplace tool." That summary fits the survey’s broader message. AI is changing who, or what, handles specific work tasks, but the evidence presented here points more toward redistribution than full automation.
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. A recent Anthropic survey of roughly 9,700 Claude users found that about half believed AI could already handle 50 percent or more of their work, though that sample came from users of a specific AI product rather than a representative cross-section of the population.
The limits of the new survey matter as well. The results are based on self-reported data. Actual time savings and AI output quality were not measured objectively, and the tasks were selected from national employment data rather than chosen for their likelihood of being affected by AI.