Epoch AI Survey Shows 20% of US Workers Delegate to AI
A joint survey by Epoch AI and Ipsos reveals that 20 percent of US workers now delegate tasks to AI instead of human colleagues, signaling a major shift toward automated task redistribution.
The study, which surveyed 1,106 employed US adults between July 10 and 19, 2026, highlights how AI is actively replacing human collaboration for specific duties. Software development leads the adoption curve, with 57 percent of workers in the field utilizing AI, followed by data analysis at 46 percent, reading work documents at 39 percent, and record-keeping at 25 percent. However, the researchers emphasize that AI is primarily serving as partial support rather than a complete replacement. Full or near-full task completion by AI only reaches 10 percent in software development and remains below 7 percent for all other categories.
The transition from human to machine delegation is most apparent in data analysis, where 7.1 percent of respondents report that AI has taken over tasks previously handled by people. This is followed by reading work documents at 5.7 percent and record-keeping at 5.3 percent. When AI manages most or all of a task, 53 percent of workers report saving time, compared to 37 percent when the technology only provides partial assistance. Conversely, about one in six AI-assisted tasks actually takes longer to complete, which may be due to the time required to interact with the system or a shift toward higher-quality output.
In terms of integration, 66 percent of AI-generated output is used with little to no editing, while only 6 percent is accepted with absolutely zero changes. On the other end of the spectrum, 5 percent of AI output requires heavy rewriting. These findings align with previous research, such as an August 2025 Gallup survey showing 45 percent of US workers use AI on the job, and an Anthropic survey of 9,700 Claude users where half believed AI could handle at least 50 percent of their workload.
For industry practitioners, these metrics indicate that AI is functioning as a task-redistribution mechanism rather than a job-killer. Epoch AI characterized the technology as a "versatile but usually not self-sufficient workplace tool," suggesting that full automation remains rare. Developers and managers should focus on optimizing human-AI collaboration workflows, particularly by addressing the inefficiencies that cause some AI-assisted tasks to take longer, rather than expecting immediate, fully automated end-to-end solutions.
This is our own summary of reporting by The Decoder
