A new survey from Epoch AI and Ipsos has found that one in five full-time American workers say AI has already taken over parts of their job, in the latest piece of data adding fuel to the burning debate over AI automation.

The poll, which surveyed 2,000 US adults, found that about half of all respondents used AI in the past week for either personal or work reasons. Among full-time workers, 20 percent said the tech had taken over tasks they used to do themselves, while 15 percent reported AI has created new on-the-job chores they wouldn’t have taken on otherwise.

In a word, this suggests that AI displacement — where AI leads to less available work for humans — is outpacing AI augmentation, where human workers become more productive as a result of access to AI tools.

“When one in five workers say AI is already replacing parts of their job, we can start talking about labor market restructuring happening in real time,” Nichols Miailhe, an AI policy leader at the Global Policy on Artificial Intelligence, told NBC about the findings. “The fact that replacement seems to be outpacing augmentation should draw our attention: the policy window to shape how AI transforms work is probably closing faster than most governments realize.”

The findings land alongside a separate, sweeping economic survey from researchers at the Federal Reserve Bank of Chicago and several top universities, which found that economists are increasingly revising their models to account for a serious shakeup of the labor market.

How true any of this is in real economic terms is becoming an increasingly urgent question. There are plenty of reasons to doubt the AI automation narrative, like the fact that AI in the workplace is still an error-prone disaster waiting to happen, if it isn’t already — like at Amazon, where the drive to replace human workers is slowing down overall productivity.

As prominent automation-doubter and AI critic Gary Marcus recently opined, the math on AI-driven unemployment doesn’t add up. Experiments on AI automation haven’t held up over time, like at the finance tech firm Klarna, which had to hire back its humans after its 11-month stunt blew up in its face.

Instead of asking whether AI is prevalent in the workplace, the better question might be whether it’s completing tasks with the same productivity as humans —and a growing consensus among experts suggests it isn’t.