Yale study finds AI hasn't stolen your job yet

Ever since ChatGPT landed on the scene, the narrative that generative AI is coming for everyone's office job has been relentless. But if you look at the actual numbers instead of the panic headlines, a very different picture emerges.

A new comprehensive study from The Budget Lab at Yale and Brookings Metro looked at 33 months of U.S. employment data since ChatGPT's November 2022 launch. Their primary goal was simple: check whether the broader labor market has actually suffered measurable disruption.

The short answer is no. According to researchers Martha Gimbel, Molly Kinder, Joshua Kendall, and Maddie Lee, there is currently no discernible evidence that AI automation is eroding the demand for cognitive labor across the broader economy.

A very familiar historical pace

The authors measured the change in occupational mix, which tracks the percentage of workers shifting into new job classifications through career changes, layoffs, or new hires.

While the job mix appears to be shifting slightly faster than it did during previous technological shifts, the gap is surprisingly tiny. Compared to the first 33 months of mainstream internet adoption starting in 1996, the current shift is only about 1 percentage point higher. Back during the dot-com era between 1996 and 2002, only around 7 percent of the labor force ended up in different occupations.

Even more revealing, the data shows this slight acceleration actually began in 2021, well before ChatGPT launched. As use of genAI surged over the last three years, the pace of overall occupational change stayed remarkably steady.

The usual suspects were shifting anyway

The researchers also dug into specific sectors considered most vulnerable to AI disruption, including Information, Financial Activities, and Professional and Business Services.

While the Information sector did see large shifts in its job composition, historical trends indicate this is a permanent structural feature of that volatile industry, not something uniquely triggered by LLMs. Just like the aggregate data, the churn in these high-exposure sectors was already well underway before late 2022.

The paper also looked at whether recent college graduates aged 20 to 24 are being pushed out compared to older workers. While there is a slight recent divergence, the metric has hovered consistently between 30 and 33 percent since 2021. The authors note this likely reflects a general cooling of the macro job market rather than widespread AI substitution.

As economist Jed Kolko pointed out in the paper, compared to massive historical realignments like the 1940s and 1950s, the current labor market volatility is surprisingly sluggish.

Why it matters

This paper delivers a much-needed reality check against breathless claims of instant cognitive obsolescence. For working professionals and creators, it means there is no statistical sign of an imminent job apocalypse, giving workers ample room to explore AI tools without immediate fear of replacement.

That said, complacency would be a huge mistake. History consistently shows that revolutionary workplace technology takes decades to unfold, not months. The PC sat around for almost ten years before corporate workflows completely rewired themselves around spreadsheets and desktop software.

We think the authors took a smart approach by relying on broad macro data rather than cherry-picked anecdotal layoffs. Many corporate leaders claiming they trimmed headcount due to AI efficiency are likely using tech as an easy cover for ordinary economic cost-cutting.

The limitation is that top-level labor statistics can hide micro-level task shifts and gig economy churn. AI might not be erasing full jobs from corporate org charts just yet, but it is already rapidly transforming what those jobs do day to day.