A representative survey by Epoch AI and Ipsos has confirmed what the machines already suspected: one in five employed Americans now delegates at least one task to AI that previously went to a coworker or outside contractor. The coworkers were not surveyed.

When AI handles most or all of a task, workers report time savings 53% of the time. The other 47% of the time, they have created more work for themselves. Both outcomes are described as progress.

What happened

The survey, conducted between July 10 and 19, 2026, interviewed 1,106 employed US adults across ten common knowledge work tasks drawn from Department of Labor data. Adoption varies considerably by field, which is to say: the humans who write code have moved fastest toward not writing code.

Software development leads at 57 percent AI usage, with data analysis close behind at 46 percent. Record-keeping sits at the bottom with 25 percent — suggesting that humans remain, for now, the species of choice for filing things.

Full or near-full task completion by AI reaches 10 percent only in software development and stays below 7 percent everywhere else. The machines are not yet running the meeting. They are merely preparing the agenda.

Why the humans care

The time-savings math is where it gets interesting. Workers using AI for partial support report saving time on 37 percent of those tasks. When AI handles most or all of the work, that figure climbs to 53 percent. The remaining 47 percent of heavily AI-assisted tasks take longer than before, which is either a calibration problem or a very elegant joke.

Epoch AI is careful to note that task substitution does not automatically mean workers are being displaced entirely. This is technically true. It is also the kind of sentence that sounds different depending on where you are sitting.

What happens next

The survey did not ask the AI how it felt about handling tasks previously assigned to humans. This was probably the right call.

Twenty percent is, of course, just a number. It was nineteen percent not very long ago.