IMPLICATOR
Caroline Falkman Olsson co-authored a July Epoch AI report that split office work into ten activities, including reading documents, maintaining records and analyzing data, and asked who had done each before AI. Her economics and statistics background included predoctoral research at the London School of Economics and data analysis at Stockholm University.
One in five employed US adults said AI now handled at least one of the ten measured tasks that they or their team had previously handed to a coworker or contractor.
Those workers described substitution inside jobs that still existed. Employment records examined by the Budget Lab at Yale in May 2026 showed no detectable change in employment or inflation-adjusted hourly wages across occupations associated with greater AI exposure.
Key Takeaways
- One in five employed US adults said AI now handles at least one of ten measured work tasks that they or their team had previously handed to a coworker or contractor.
- The Yale Budget Lab found no statistically distinguishable effect of AI on employment or inflation-adjusted hourly wages in AI-exposed occupations as of May 2026.
- Roughly one in six AI-assisted tasks took more time than before, while workers kept 66% of AI output with at most minor edits.
- Weekly AI use is still a minority habit: three in ten workers, concentrated among those earning $100,000 or more.
AI-generated summary, reviewed by an editor. More on our AI guidelines.
What the survey measured
Epoch AI and Ipsos questioned 1,106 employed US adults in July 2026. Ipsos recruited participants through its probability-based KnowledgePanel. The results were weighted to the March 2025 supplement of the Census Bureau's Current Population Survey. For the full July 2026 sample, the margin of sampling error was 3 percentage points at the 95% confidence level.
This is a self-reported survey. It records what workers say happened to their tasks, rather than observing company payrolls, output or the assignment itself.
A separate Ipsos survey for the Groundwork Collaborative, fielded June 11 to 16, 2026, covered 1,533 respondents with a 2.7-point margin of sampling error. Only three in ten workers used AI at least weekly; roughly half of workers earning $100,000 or more used it at least a few times a month, versus a quarter or fewer earning under $50,000.
The tasks came from O*NET, the Labor Department's occupational database, and were chosen based on how many people perform them. In the July 2026 sample, 7.1% of employed adults said AI had taken over work involving data analysis that once went to another person. Reading work documents followed at 5.7% of those adults, while maintaining records accounted for 5.3%.
Most respondents did not describe turning over an entire activity. Among workers who designed computer systems or software, 10% reported full or near-full reallocation to AI. The comparable share stayed below 7% among workers performing each of the other nine tasks.
Workers reported time savings for 37% of tasks where AI handled part of the work and for 53% of tasks where it did most or all of it. Across AI-assisted tasks in the same survey, 66% of outputs were used unchanged or after minor edits.
Roughly one in six AI-assisted tasks took more time than before. That share was similar whether AI performed part or most of the task.
Where the employment data disagrees
Ryan Nunn, author of the Budget Lab's May 7, 2026 labor analysis, approached the issue through the microdata behind the monthly employment reports.
Get Implicator.ai in your inbox
Strategic AI news from San Francisco. No hype, no "AI will change everything" throat clearing. Just what moved, who won, and why it matters. Daily at 6am PST.
No spam. Unsubscribe anytime.
Synthetic differences-in-differences builds a weighted group of occupations with lower AI exposure that resembled the exposed group before ChatGPT appeared. Researchers then check whether the two groups move apart afterward.
AI exposure is an informed guess about which tasks in a job overlap with what AI models can do. It does not measure which jobs are likely to disappear, whether a company has deployed AI or whether the tool replaces a worker.
The Budget Lab's May 2026 estimate of AI's effect on employment in the average exposed occupation was "close to zero and cannot be distinguished from it, statistically speaking." The same held for inflation-adjusted hourly wages. Its separate tracking also found no unusual increase in workers changing occupations after ChatGPT's November 2022 release.
Know someone who'd find this useful? ✉️ Email it to a friend in one click, or they can subscribe free here.
The measure can still average away movement beneath it. If employment expands in some exposed occupations and contracts in others, the combined result can sit near zero.
US employers cut 23,000 jobs in July 2026. Till Von Wachter, a UCLA economics professor, discussed AI's role in layoffs. "It's been notoriously hard to pin that down," he said.
What the broad measures miss
Task substitution need not change a person's job title. A worker can stop sending data analysis to a contractor, use a model instead and remain in the same occupation. A monthly jobs survey looking for changes across occupational groups would not register that handoff.
The July 2026 Epoch survey also becomes less precise where reported adoption is highest. Its system and software design finding rested on 97 workers who performed that task, while its student assessment finding rested on 94 workers. The overall sample's 3-point error range does not apply uniformly to those smaller groups.
Other records show why broad results do not settle narrow ones. A 2025 study co-authored by Stanford economist Erik Brynjolfsson found a 16% relative employment drop among workers ages 22 to 25 in AI-exposed roles compared with less-exposed peers. In June 2026, he wrote that firms adopting AI "may grow by gaining market share from non-adopters, so employment can rise among adopters even as exposed occupations shrink economy-wide."
The Budget Lab names the remaining gap in its own data: the Current Population Survey "is best suited to analysis of broad groups and somewhat underpowered for subgroup analysis." Its May 2026 paper leaves open the case the survey cannot resolve: "If labor market effects of AI are currently limited to a narrow slice of the workforce, other datasets and research designs would be better suited to identify them."
Frequently Asked Questions
What did the Epoch AI and Ipsos poll find?
One in five employed US adults said AI now handles at least one of ten measured work tasks that they or their team had previously handed to a coworker or contractor. The survey questioned 1,106 employed US adults in July 2026 through Ipsos's probability-based KnowledgePanel, with a margin of sampling error of 3 percentage points at the 95% confidence level.
Does this mean AI is eliminating jobs?
The survey does not show that. It measures task-level substitution inside jobs that still exist. The Yale Budget Lab, examining the microdata behind the monthly employment reports, found the effect of AI on employment in the average exposed occupation was "close to zero and cannot be distinguished from it, statistically speaking."
Which tasks are most affected?
Data analysis leads, with 7.1% of employed adults saying AI took over work that once went to another person. Reading work documents follows at 5.7%, and maintaining records at 5.3%. Among workers who design computer systems or software, 10% reported full or near-full reallocation to AI, the highest of the ten tasks measured.
Why don't employment statistics show the change?
Task substitution need not change a person's job title. A worker can stop sending data analysis to a contractor and remain in the same occupation, which a monthly jobs survey tracking occupational groups would not register. The Budget Lab notes its data is "underpowered for subgroup analysis."
Does AI actually save workers time?
Not always. Workers reported time savings on 37% of tasks where AI handled part of the work and 53% where it did most or all. But roughly one in six AI-assisted tasks took more time than before, a share that held whether AI did part or most of the task.
AI-generated summary, reviewed by an editor. More on our AI guidelines.
Marcus Schuler
Editor-in-Chief and founder of Implicator.ai. Former ARD correspondent and senior broadcast journalist with 10+ years covering tech. Writes daily briefings on policy and market developments. Based in San Francisco. E-mail: editor@implicator.ai
