A representative survey by Epoch AI found that 20% of employed Americans hand off at least one task to AI that a human used to do, and they generally accept AI output with little to no editing. This indicates a growing reliance on AI in the workplace, potentially reshaping how tasks are assigned and completed.
SynthePulse Insight · AI deep reading
Version 1 · 1 source
A new survey by Epoch AI and Ipsos shows that 20% of employed U.S. adults have delegated at least one task previously done by a colleague or contractor to AI. But a closer look at the data reveals that AI is more often a 'partial helper' than a 'fully automated substitute,' and it doesn't always save time.
From July 10 to 19, 2026, Epoch AI and polling firm Ipsos conducted a representative survey of 1,106 employed U.S. adults. Respondents were asked about ten common work tasks, selected from U.S. Department of Labor data to reflect typical knowledge work. The results show that 20% of respondents said at least one task previously done by a colleague or contractor is now handled by AI.
Epoch AI describes AI as a 'versatile but often not self-sufficient workplace tool' and emphasizes that the data points more to a reallocation of tasks between humans and AI rather than full automation of entire jobs.
Among the ten tasks, AI usage rates vary significantly. The highest usage is in computer systems and software development at 57%, followed by data analysis at 46%, reading work documents at 39%, and record keeping at the bottom with 25%.
However, using AI does not mean the machine takes over the entire task. Most respondents describe AI as providing partial support. Only in software development does AI complete all or nearly all of the task, at 10%, while all other tasks are below 7%.
The shift from human to machine work is most pronounced in data analysis: 7.1% of respondents said AI has taken over tasks previously handled by humans. Reading work documents (5.7%) and record keeping (5.3%) follow closely. Epoch AI emphasizes that such task substitution does not automatically imply that workers are fully replaced.
The survey shows a correlation between the level of AI involvement and reported time savings. When AI handles only part of a task, 37% of cases report time savings; when AI completes most or all of the work, this figure rises to 53%.
However, researchers note that it cannot be determined from the data whether more AI use leads to faster completion, or whether workers are more likely to use AI when they want to save time.
Additionally, about one in six AI-assisted tasks actually take longer than before. Researchers speculate that this could be because interacting with AI itself consumes time, or because workers use the saved time to complete tasks more thoroughly or with higher quality.
Regarding how workers handle AI output, 66% of AI outputs are used as-is or with minor edits; only 6% are used completely unmodified, while 5% are heavily reworked or mostly rewritten.
Researchers point out that low editing effort does not directly measure the quality of AI output, and they found no consistent association between reported time savings and the amount of editing.
These findings align with previous surveys. A Gallup survey from August 2025 found that 45% of U.S. workers use AI at work, but only 10% use it daily. An Anthropic survey of about 9,700 Claude users showed that roughly half of respondents believe AI can handle 50% or more of their work. However, Anthropic's sample consists of users of a specific AI product and is not a population-representative sample.
This report is based on THE DECODER's coverage of an Epoch AI survey conducted with Ipsos, with a sample of 1,106 employed U.S. adults. All data comes from self-reports in the survey and has not been objectively measured. Epoch AI is a first-party source, but this report is a secondary account, so specific figures and conclusions should be treated as source claims rather than confirmed facts.
AI is taking on more tasks in the U.S. workplace that were previously done by humans, but in most cases it provides partial support rather than full replacement. While AI use is associated with time savings, this is not always the case, and output quality has not been objectively verified. The survey suggests that AI is more of a task-reallocation tool than a fully automated solution.
Primary report
Primary source