- Published on
One in five US workers now delegates tasks to AI instead of colleagues, survey finds
- Authors
- Name
- DeepHorus
One in five employed Americans has handed at least one work task to artificial intelligence that used to be done by a coworker or outside contractor, according to a new survey from Epoch AI and polling firm Ipsos. The finding signals a fundamental shift in how knowledge work is allocated, as AI becomes a regular member of the workforce rather than an experimental tool. The survey, which polled 1,106 US adults between July 10 and 19, 2026, asked about ten common tasks drawn from Department of Labor data.
Adoption is uneven across job functions. Software development leads the pack, with 57 percent of workers in that field using AI for at least part of the job. Data analysis follows at 46 percent, and reading work documents at 39 percent. Record-keeping sits at the bottom, with just 25 percent of affected workers turning to AI. These figures show that AI has penetrated deeply into technical roles while remaining less prevalent in routine clerical work.
In most cases, however, AI is not taking over entire tasks. Workers described AI as partial support for the vast majority of their work. Full or near-full completion by AI hits just 10 percent in software development and stays below 7 percent in every other category. The most striking shift from human to machine appears in data analysis, where 7.1 percent of respondents say AI now handles tasks previously done by people.
The survey also explored time savings. When AI handles only part of a job, workers report saving time on 37 percent of those tasks. When AI does most or all of the work, that figure jumps to 53 percent. But causality is unclear, the researchers caution. It could be that heavier AI use makes workers faster, or that workers choose to lean on AI more when they want to free up time.
AI doesn't always speed things up. Roughly one in six AI-assisted tasks now takes longer than before, suggesting that interacting with the model can be time-consuming, or that workers use freed-up capacity to perform tasks more thoroughly. The quality of AI output appears generally acceptable to users: 66 percent of output is used unchanged or with only minor tweaks, while just 5 percent is heavily reworked.
Epoch AI sums up the situation by calling AI a "versatile but usually not self-sufficient workplace tool." The data points to a redistribution of tasks between humans and AI rather than mass automation of entire jobs. This nuance matters for policymakers and business leaders worried about displacement. So far, the evidence suggests AI is augmenting roles, not eliminating them, but the trend bears watching.
Looking ahead, the adoption curve is likely to continue rising as models improve and workers become more comfortable with delegation. The fact that even record-keeping, the least-affected task, shows 25 percent adoption indicates that AI is becoming ubiquitous. Yet the lack of objective measurement and the significant minority of tasks that take longer suggest that we are still in the early stages of figuring out how to integrate AI effectively. Expect more emphasis on training and workflow redesign in the coming years.
