As of 2026, the most careful research from Yale's Budget Lab finds no clear sign that AI has reshaped the overall labor market. Unemployment has drifted up, but it has done so for ordinary economic reasons, not a robot takeover.
That said, not yet at the macro level is not the same as nothing is happening. There are real, specific signals worth taking seriously, especially for young workers entering AI-exposed fields. The real story is pressure concentrated in particular places rather than displacement across the board, and understanding that difference is the whole point.
Is AI Taking Jobs at the Macro Level Yet?
When people ask is AI taking jobs, they usually mean something broad. Is the overall number of jobs shrinking because of AI? Here the evidence is surprisingly clear, and surprisingly calm. Yale's Budget Lab tracks how quickly the mix of occupations in the US economy is changing. In its October 2025 analysis, the pace of occupational change since ChatGPT was only about 1 percentage point higher than during the early internet era of the late 1990s, and much of that shift actually began in 2021, before ChatGPT existed.
A follow-up analysis in May 2026 went further, using a more rigorous method to compare AI-exposed occupations against everything else. The estimated employment effect was so close to zero that it could not be statistically distinguished from zero. Unemployment did rise over this period, from 3.4 percent in April 2023 to 4.3 percent in March 2026, but that softening shows up across exposed and unexposed jobs alike, which is exactly what you would not expect if AI were the cause.
What Does the Occupational Data Actually Show?
It helps to separate three questions that often get blurred together. What could AI do, what are people using it for, and what has actually happened to jobs. Each has a different answer, and mixing them up is where most bad takes come from.
On potential, the landmark study is Eloundou and colleagues, published in Science in 2024. They estimated that roughly 1.8 percent of jobs could have more than half their tasks affected by LLMs with simple interfaces and general training, rising to just over 46 percent of jobs once you account for the complementary software being built around those models. That is a statement about exposure, not about jobs lost. The gap between broad exposure and no discernible macro disruption is the single most important thing to understand here. Capability is not adoption, and adoption is not replacement.
Why Are Young Workers the Real Warning Sign?
In November 2025, Stanford's Digital Economy Lab published Canaries in the Coal Mine, an analysis by Erik Brynjolfsson and colleagues using payroll records from the largest payroll provider in the US. They found that early-career workers ages 22 to 25 in the most AI- exposed occupations experienced a 16 percent relative decline in employment since generative AI became widespread. Over the same window, older and more experienced workers in those same fields kept growing.
Two cautions matter. First, the authors frame this as consistent with a hypothesis, not proof, since interest rates and post- pandemic hiring corrections are still being untangled. Second, relative decline means young workers in exposed roles fell behind their peers, not that a huge share of the workforce vanished. Still, it is a genuine signal, and it points at entry-level tasks, the routine, learnable work that both new hires and AI tools are good at.
If AI Is So Capable, Why Has It Not Shown Up in the Numbers?
If nearly half of all jobs could have most of their tasks affected, why is the macro data so quiet? Because using AI and replacing a job are very different things. The Federal Reserve Bank of St. Louis found that as of August 2025, 37.4 percent of workers ages 18 to 64 used generative AI at work, up from 33.3 percent a year earlier. But those users spent only about 5.7 percent of their work hours actually using it. People are adopting AI for slices of their day, not handing over whole jobs.
Most of that usage leans toward helping people rather than replacing them. Anthropic's Economic Index (2025), which studies how people use Claude, found that 57 percent of usage looked like augmentation, where the tool helps a person do their work, versus 43 percent that looked like automation. Right now AI absorbs individual tasks while leaving the occupations around them intact, which is exactly why a task-level view matters more than a scary job-title headline. Our tool at aijobriskcheck.com is built on that distinction. It maps your actual daily tasks to occupational data and scores each one against current AI capability.
What Did the 2026 Sources Actually Measure?
These sources do not contradict each other. They answer different questions. Read together, they show broad exposure, growing adoption, usage concentrated in parts of the workday, and no clean fingerprint of mass displacement in the aggregate, alongside one sharp signal among the youngest exposed workers.
What Should You Actually Do About It?
The wrong response is either nothing is happening, relax, or everything is doomed, panic. The right response is specific and calm. Look at your own tasks, not your job title. Two people with the same title can have very different exposure depending on how they spend their day. Identify which of your recurring tasks are the routine, well-documented, text-and-data ones that current tools handle well, and which depend on judgment, relationships, physical presence, or accountability that AI cannot hold.
Then shift deliberately. If you are early in your career, the young-worker signal is a reason to build the skills that sit above the automatable layer, including reviewing and directing AI output, owning outcomes, and handling the messy human parts of a role. Think of it less as outrunning a machine and more as knowing exactly where you stand so you can move on purpose.
Get your free AI Job Risk Score. Tell us your job title and how you actually spend your time, and we will show you which of your tasks are exposed today. Free. 60 seconds. No sign-up required.
Get MY AI Risk Score NowFrequently Asked Questions
Are young workers being hit by AI?
Yes, relative to peers. Stanford found a 16 percent relative employment decline for ages 22 to 25 in exposed roles. Relative means they fell behind less-exposed peers, not that jobs vanished outright. Entry-level routine tasks carry the risk.
What percentage of jobs will AI replace?
No credible source gives a firm number. Exposure estimates are the closest thing: about 80 percent of workers have at least 10 percent of tasks exposed to large language models. Treat any confident replacement percentage as a red flag.
Does this mean my job is safe?
No guarantee exists for any specific job. Start by listing your ten most common weekly tasks. Routine, well-documented text and data work is most exposed; work requiring judgment, relationships, or accountability is least. Build skills around the second group.
Sources
- The Budget Lab at Yale, Evaluating the Impact of AI on the Labor Market, October 2025. Read the analysis.
- The Budget Lab at Yale, AI Is Probably Not (Yet) the Reason for Labor Market Weakening, May 2026. Read the follow-up.
- Stanford Digital Economy Lab, Canaries in the Coal Mine (Nov 2025).
- Federal Reserve Bank of St. Louis, The State of Generative AI Adoption in 2025, November 2025. Read the report.
- Anthropic Economic Index, 2025. Read the index.
- Eloundou, Manning, Mishkin, and Rock, GPTs are GPTs, Science, 2024. View in Science.