Is AI coming for your precious (job)?
The headlines say Gollum is at the door. The data says it's more complicated than that.
Nowadays it feels like our job is the One Ring, and AI is Gollum, crazed and creeping after our precious job.
Every headline has him reaching for it again, every week another voice swearing he’s already at the door. But screaming he’s at the door is not the same as seeing him there. So let’s look at how close Gollum actually is.
A Fellowship of Headlines
For sure, you’ve seen the headlines claiming that AI could wipe out half of all entry-level white-collar jobs within five years and push unemployment to 10–20%. This isn’t an invented quote (I checked). It comes from Anthropic CEO Dario Amodei, who shared this forecast in an interview in May 20251.
Before you let these numbers ruin your day, I want to clarify two things.
First, notice how these headlines are often framed: 50% of entry-level jobs will disappear. Add a precise number to a claim, and it suddenly feels less like a guess and more like a measurement—even when it isn’t. Humans are surprisingly easy to impress with decimals and percentages2.
Second, I went looking for the source behind Dario’s prediction, and I found nothing to back up that specific number. As a quantitative researcher, I have a hard time accepting precise figures without a source, a model, or at least some estimation behind them. AI is still a very new phenomenon, so I can tolerate a bit of educated guesswork (but even educated guesses should come with some kind of evidence).
Maybe he’s right. Maybe he’s wrong. The future has a habit of humbling everyone. But the confidence of the headline seems to go a bit further than the evidence available today. It felt as if the discussion took a small detour from measuring what AI is doing now to confidently forecasting what it will do next. Or, in Tolkien terms, we somehow went from a walk through the Shire to a journey to Mordor in a single paragraph.
However, pay attention, little hobbits: Dario is not alone in Mordor.
Elon Musk said that "probably none of us will have a job." Sam Altman warned that entire job categories would be "totally, totally gone." Mustafa Suleyman, head of Microsoft AI, argued that most white-collar tasks could be fully automated within 12 to 18 months. Every one of those statements became a headline3.
But let’s be fair, not everyone in Mordor is Sauron. While some AI leaders have made dramatic predictions about mass unemployment, others have been pushing back4. Jensen Huang, CEO of Nvidia, recently criticized executives who blame layoffs on AI, arguing that many are using the technology as a convenient excuse. Demis Hassabis, CEO of Google DeepMind, has emphasized that AI will also create new jobs and that humans are remarkably adaptable.
Perhaps most interestingly, some of the same voices that filled the headlines with fire are now turning the temperature down a little. Sam Altman recently admitted that AI has eliminated far fewer entry-level jobs than he expected5. At the same time, Dario Amodei has increasingly discussed how AI could make workers more productive rather than simply replace them6.
That doesn’t necessarily mean the danger has passed, but it does suggest that headlines may not be the best place to build your plans—or your fears. If your expectations about the future swing with every dramatic prediction, you’ll spend more time chasing headlines than understanding what is actually happening.
The Two Papers
If headlines are misleading, let’s turn to science. Science never fails (said the scientist writing this article).
The good news is that, even though AI is a relatively recent phenomenon, we already have some strong and robust research to work with. The bad news is that the road from the Shire to Mordor may indeed have been traveled by at least some hobbits.
The study that caught everyone’s attention is called Canaries in the Coal Mine7, and the title alone hints at the researchers’ conclusions.
The researchers from Stanford behind this study had a straightforward goal: figure out whether AI is actually affecting employment, and if so, for whom. To do that, they needed two things. First, a measure of employment and for that they used real payroll data (which is a much more reliable picture than surveys). Second, they needed a way to measure how exposed each job is to AI (meaning how much of the work in that role could plausibly be done by an AI system today — and worth keeping in mind that this study was published in November 2025, so what counts as “exposed” may look quite different in a year, given how fast AI is moving). They built that exposure measure by scoring every occupation twice: once using a GPT-4 based index that breaks each job down task by task and asks how susceptible each task is to AI, and once using real-world data from millions of actual Claude conversations to see which jobs people are already turning to AI to help with. Then they matched both scores to the payroll data and watched what happened over time.
What they found was not good news for young Frodo and Sam.
Workers aged 22 to 25 in the most AI-exposed occupations saw roughly a 13% relative decline in employment since late 2022 (that’s not a small number, it means that for every 100 young workers in those jobs before, there are now around 87). Older workers in those exact same jobs? Stable, or even slightly up.
Here is where the article title earns its place. Miners carried canaries underground because the birds were the first to feel the danger. In some ways, the AI effect described in this paper is less like an orc attack and more like the influence of the Ring itself. For example, during a recession, everyone knows something is wrong: layoffs are announced, unemployment rises, and headlines follow. What these researchers describe is quieter. Instead of workers losing jobs, an entire generation may slowly stop getting hired into them. Because the change is so gradual, very little is said about it and even less is planned for its consequences. Which raises a question that nobody seems to be asking loudly enough: what does the future of the workforce actually look like if the bottom rung quietly disappears?
While this is quite worrying, especially for new graduates entering the labor market every year, I have to raise a flag and make something very clear: this study does not prove that AI caused the decline in employment.
In plain English, the Stanford researchers observed that two things happened at the same time: AI became more widespread, and young workers in AI-exposed occupations started getting hired less. But when two things happen together, that does not automatically mean one caused the other. Ice cream sales and sunburns both rise every summer, but ice cream is not causing the sunburns. The real culprit is the hot weather.
To their credit, the researchers did try to rule out alternative explanations. Was this driven only by tech firms? It wasn’t. Was it simply the result of changes in interest rates? It wasn’t. Was it explained by occupations that could easily be performed remotely? It wasn’t.
But let’s go back to our hot-weather culprit story. If AI is not guilty, what else could be happening? Why are young workers getting hired less? Another team of researchers came up with a completely different suspect: remote work8. Yes, remote work (not the answer you were expecting, right?).
The other team of researchers used a different approach. First, to measure employment, they used new hire records and job postings across the US, UK, Canada and Australia. To measure their two suspects, they scored every occupation on both AI exposure (using the same task-by-task index as the Stanford team) and remote work exposure (using an index that scores how feasible each job is to do from home, based on the nature of its tasks). Here is where it gets interesting, those two scores turned out to be almost identical lists. The occupations most exposed to AI are almost exactly the same occupations most likely to be done remotely. This is where the path to Mordor gets a little tricky.
For you to understand, imagine you walk into a room and find the floor covered in water. The window is open, and it is raining outside, but there is also a leaking pipe. At first glance, both look like plausible explanations. The challenge is figuring out how much of the water each one actually caused.
When the researchers accounted for both AI and remote work at the same time, remote work turned out to explain the decline in hiring. Once remote work was taken into account, AI no longer added much to the explanation. In our flooded-room story, the leaking pipe (remote work) was enough to explain the water on the floor (lower hiring). The rain (AI) may still have been there, but there was no evidence that it was the reason the room was flooded.
So is it AI or remote work?
Well, as a social scientist, I have to admit something that sometimes makes people uncomfortable: social science rarely gives us neat, final answers. Humans and societies are messy, and context matters—a lot.
Before declaring a winner, take a look at the table below. The differences between the two studies are not just in the results (they are also in the context, the data,...):
I don’t want to get lost in the methodological weeds here, but I do want to make sure you understand something important: both of these are valid and robust studies, yet they arrive at different conclusions. That is precisely why social phenomena are so difficult to pin down.
It is also why policy debates can become messy. Politicians (and sometimes researchers too—I won’t lie to you) may highlight the findings that best support their preferred narrative rather than considering the full body of evidence.
But you cannot afford to be fooled. This is not just about policy or research. It is about your future and how you navigate the AI job apocalypse that is supposedly upon us. So let’s leave research for a moment and look at what the data is actually showing us.
The Return of the Data
The data is kind of messy, right? Bad news, my nerd friend, it is like that most of the time.
Data is just a bunch of different pieces of information, each showing one side of the story rather than a clean and complete panorama of reality.
So what do you actually do with all this?
You do the only thing that survives contact with all the limitations: you stop looking for one source and start triangulating between them. You hold the scary headline next to the scientific paper next to the official statistics, and you treat the disagreement between them as information rather than noise.
That is actually why critical thinking has become one of the most talked-about skills in the age of AI. Getting an answer has never been easier, but a reliable answer is a different story.
Your AI oracle can give you a quick response in seconds, but it may miss context, overlook alternative explanations, or simply reflect the limitations of the information it was trained on. So instead of asking, “What’s the answer?”, try asking, “What are the competing explanations? What evidence supports each one? What evidence goes against them? And what would change my mind?”
If you want to get closer to reality or the truth (if it exists), the best way is still the old-fashioned way: compare sources, challenge assumptions, and let the evidence argue with itself before you make up your mind.
So, are you going to keep your precious (job)?
The most honest answer is that I don’t know.
Although by triangulating different sources and looking at previous technological transformations, it seems very likely that the job market will change. There will be winners and losers. How many of each? That part I will leave to the headlines.
What I do think is that Gollum seems more interested in some rings than others. So far, the evidence points most strongly toward two characteristics: tasks that are structured, repetitive, and easy to codify, and tasks that require relatively little judgment. Because many of these activities have traditionally been assigned to junior workers in some fields, this pattern is starting to show up both in the scientific literature and in the headlines.
What I am coming to the conclusion, based on all this research and by combining many different sources of information, is that the most promising strategy seems to be moving in the exact opposite direction of what AI does best. If AI excels at following clear instructions, pattern matching, and producing answers from existing information, then the skills that become more valuable are those that involve judgment, creativity, dealing with ambiguity, communication, and understanding people.
According to the World Economic Forum’s latest Future of Jobs report9, the skills expected to grow most in importance include analytical thinking, creative thinking, resilience, flexibility, leadership, technological literacy, and AI literacy. In other words, the future may belong neither to the people who ignore AI nor to those who blindly follow it, but to those who can work with it while contributing to the things it still struggles to do well.
In the end, the picture becomes much less apocalyptic (and much more interesting) once you stop relying on a single source and start connecting the dots between many of them.
I hope you’ll join me on this quest for more data points as we try to separate signal from noise and figure out what the future really has in store for us.
VandeHei, J., & Allen, M. (2025, May 28). Behind the curtain: A white-collar bloodbath. Axios. https://www.axios.com/2025/05/28/ai-jobs-white-collar-unemployment-anthropic
Numbers are persuasive—if used in moderation. (2025, May 29). Scientific American. https://www.scientificamerican.com/article/numbers-are-persuasive-if-used-in-moderation/
Hartmans, A. (2024, May 23). Elon Musk says AI will take all our jobs. CNN Business. https://edition.cnn.com/2024/05/23/tech/elon-musk-ai-your-job
Altman (”totally, totally gone”):
Wiggers, K. (2025, July 24). OpenAI CEO Sam Altman warns AI will wipe entire job categories off the map. Tom’s Guide. https://www.tomsguide.com/ai/openai-ceo-sam-altman-warns-ai-will-wipe-entire-job-categories-off-the-map
Suleyman:
Wieczner, J. (2026, February 13). Microsoft AI chief gives it 18 months — for all white-collar work to be automated by AI. Fortune. https://fortune.com/2026/02/13/when-will-ai-kill-white-collar-office-jobs-18-months-microsoft-mustafa-suleyman/
Shibu, S. (2026, May 27). CEOs are blaming AI for layoffs. Nvidia’s Jensen Huang says that’s a ‘lazy’ excuse. Entrepreneur. https://www.entrepreneur.com/business-news/nvidias-jensen-huang-says-that-ai-is-a-lazy-excuse-for-layoffs
Langley, H. (2025, June 3). Google’s Demis Hassabis says AI will create new ‘very valuable jobs.’ Business Insider. https://www.businessinsider.com/demis-hassabis-google-deemind-study-future-jobs-ai-2025-6
Reints, R. (2026, May 26). Sam Altman says AI ‘jobs apocalypse’ probably won’t happen. Time. https://time.com/article/2026/05/26/sam-altman-ai-job-losses-openAI-/
Lichtenberg, N. (2026, May 5). Dario Amodei spent last year warning of an AI white-collar bloodbath. Now he’s changing the narrative. Fortune. https://fortune.com/2026/05/05/dario-amodei-jevons-paradox-will-ai-wipe-out-white-collar-jobs/
Brynjolfsson, E., Chandar, B., & Chen, R. (2025). Canaries in the coal mine? Six facts about the recent employment effects of artificial intelligence. Stanford Digital Economy Lab. https://digitaleconomy.stanford.edu/publications/canaries-in-the-coal-mine/
Lambert, P. J., & Schindler, Y. (2026). The broken ladder: AI, remote work, and early-career hiring. SSRN. https://doi.org/10.2139/ssrn.6787638
World Economic Forum. (2025). The future of jobs report 2025. World Economic Forum. https://www.weforum.org/publications/the-future-of-jobs-report-2025/