Figure 2: Theoretical capability and observed exposure by occupational category. Share of job tasks that LLMs could theoretically perform (blue area) and our own job coverage measure derived from usage data (red area).
Posted on 04/28/2026 8:00:23 AM PDT by ProtectOurFreedom
Anthropic just published a study mapping exactly which jobs its own AI is replacing right now.
The workers most at risk are not who anyone expected. They are older. They are more educated. They earn 47% more than average. And they are nearly four times more likely to hold a graduate degree than the workers AI is not touching.
The argument is straightforward. Anthropic built a new metric called "observed exposure." Not what AI could theoretically do. What it is actually doing right now in professional settings, measured against millions of real Claude conversations from enterprise users.
For computer and math workers, AI is theoretically capable of handling 94% of their tasks. It is currently handling 33% of them. For office and administrative roles, theoretical capability is 90%. Current observed usage is 40%. The gap between what AI can do and what it is already doing is enormous. The researchers are explicit about what comes next. As capabilities improve and adoption deepens, the red area grows to fill the blue.
The demographic finding is what makes the paper uncomfortable. The most AI-exposed workers earn 47% more on average than the least exposed group. They are more likely to be female. They are more likely to be college educated. This is not a story about warehouse workers or truck drivers. It is a story about lawyers, financial analysts, market researchers, and software developers. The exact group whose education was supposed to insulate them.
Computer programmers showed the highest observed AI exposure at 74.5%. Customer service representatives at 70.1%. Data entry keyers at 67.1%. Medical record specialists at 66.7%. Market research analysts and marketing specialists at 64.8%. These are not predictions. These are measurements of work that is already happening on AI platforms right now.
Then there is the pipeline finding nobody is talking about loudly enough. Anthropic's researchers found a 14% decline in the job-finding rate for workers aged 22 to 25 in highly exposed occupations since ChatGPT launched. No comparable effect for workers over 25. Entry-level roles were never just jobs. They were the training ground where junior analysts became senior analysts, where junior lawyers learned how arguments hold together. If that layer disappears, nobody has answered the question of where the next generation of senior professionals comes from.
The detail buried in the paper that most coverage missed: 30% of American workers have zero AI exposure at all. Cooks. Mechanics. Bartenders. Dishwashers. The technology reshaping professional careers is completely irrelevant to roughly a third of the workforce. The divide is no longer between high skill and low skill. It is between presence and absence.
Figure 2: Theoretical capability and observed exposure by occupational category. Share of job tasks that LLMs could theoretically perform (blue area) and our own job coverage measure derived from usage data (red area).
The company publishing this study is the same company selling the AI doing the replacing. Anthropic had every commercial incentive to soften these findings. They published them anyway.
If you spent four years and $200,000 on a degree to land a white collar career, the company that builds Claude just confirmed your job is more exposed than the bartender pouring drinks at your graduation party.
Figure 3: Most exposed occupations. Top ten most exposed occupations using our task coverage measure.
Original Source Report by Anthropic: Labor market impacts of AI: A new measure and early evidence
Do you see that playing out any other way?
Yes.
That's the static view–AI does the work previously done by staff, fewer people required. The boss picks up any slack remaining.
But that's not how it plays out in genuinely competitive fields.
AI doesn't just help a securities analyst do his existing job faster. It opens up information horizons that didn't exist before. He can now track operational metrics, supply chain signals, and industry dynamics at a level of granularity that was simply out of reach. That's not the same work done more efficiently–it's a fundamentally larger canvas. The firm that cuts its junior staff to pocket the savings will be outcompeted by the firm that redeploys that capacity into the new territory AI has made accessible.
Same in law. AI doesn't just help a defense attorney process the same evidence faster. It surfaces connections across case law, forensic data, and precedent that no junior associate could have found in time. The attorney who uses AI to shrink his team will eventually face one who used it to build a deeper, more comprehensive case.
The zero-sum assumption treats the work as fixed. In competitive professional fields, AI makes the work larger. The firms and attorneys and analysts who grasp that will eat the lunch of the ones who just used it to cut headcount.
As an aside, I am a mountain biker and recently bought a weighted ruckpack (GoRuck) so I can hike my biking trails. Have you looked into it?
I've only been at it for a couple of months, but am increasingly confident it is going to be great addition to my fitness routine.
You are describing the well-known “Luddite fallacy” (also tied to the “lump of labor” fallacy). The fallacy is the mistaken belief that technological progress and automation permanently destroy more jobs than they create, leading to long-term structural unemployment. It assumes there’s a fixed “lump” of work to be done in the economy—so if machines take over tasks, humans must lose out overall.
We shall see. I believe that is probably true out maybe 10 years, but, beyond that, all bets are off.
But how soon does the Law of Diminishing Returns kick in with AI?
To add insult to injury-23 yr olds had to deal with Covid their senior year in high school and their freshman year of college. They got screwed in so many important ways of growing up.
Thanks.
Funny you mention rucking. Yes, I’ve been looking at ruck weights for my pack. For the time being, I’m just adding more water bottles to my camera gear. My pack is at about 20 pounds now.
I did a hike up Rhus Ridge a few days ago — a 2.5 mile loop and 700 feet elevation gain. It’s through the hills above Silicon Valley. My app previously calculated about 530 calorie burn on that loop with the 20 pound backpack. I had a slight groin muscle pull, so backed off to my simple daypack (4.1 pounds, one water bottle) and I got 380 calories! I asked Claude “What happened?” and it immediately replied you were carrying less weight. I plugged in my 20 pound backpack and immediately got 530 calories again. ( I started writing the app in Grok and switched to the superior Claude).
It was amazing that Claude (which had helped write the program) understood the program so well it immediately said “You had less weight in your pack.”
I had no idea that 16 additional pounds over 2.5 miles / 700 feet added 150 calorie burn!
I’ve been paying a lot of attention to fitness, my weight, my body composition. I’ve lost 34 pounds in 7 months and my app has helped me a lot stay on track.
Good point. They are really getting the double-whammy, aren’t they?
You would be absolutely stunned to understand just how much AI will rock the lawyer industry. AI can set the groundwork for most cases better than any attorney. Oh yeah-it is 20 bucks per month and takes 20 seconds.
I felt terrible for my child that graduated high school in the class of 2020. Graduation was getting your diploma on the courthouse steps with 5 minutes between each graduate while all the families sat in their cars and watched.
The problem with AI is that it is frontier country. All we can do is speculate as to what's coming.
That must have brutal to watch and endure. Sorry you had to go through that.
We escaped largely unscathed. Our youngest was 29 at the time. I was retired. My wife’s job at the school district went remote (San Fran Peninsula). Newsom said he was going to shut down all traffic on the interstate highways (!) so my wife said “Screw that, I’m outta here.” She grabbed our dog, hopped into her car, and drove to our North Idaho place. She left CA for good! Had a couple of business trips back here, but became a true Idahoan where we are still immune to insanity.
I spend some time at our place in CA every winter to get more sunshine, but arriving here and still seeing mask requirements all over the place just nauseates me.
I got runner's knee in the second week from overdoing it. I took a week off and have started back at 2 miles–slowly. My knee feels fine so far.
There's something wonderful about weighted rucking. Not sure why.
I've got a 30lb plate on backlog order. But I am going to progress slowly from now on.;-)
I had a slight tear on my left meniscus (knee) about 20 years ago. It was painful. It healed up pretty good without surgery, but I always wear a tight knee brace. Hiking poles offload the knees a bit. So far, so good.
That's an interesting question. I pondered it overnight.
I'm just thinking out loud here, of course, but it occurred to me that there's a deeper point here that we might be missing.
AI, as we understand it right now, doesn't just help us process a fixed universe of information more efficiently–it expands the information universe itself. As we understand more, we find more worth understanding. Complexity breeds complexity. Human behavior, markets, institutions, and science all respond to new knowledge by generating new questions.
Think of it like a microscope that keeps getting more powerful. Diminishing returns don't kick in until either we can no longer build a better microscope, or we've seen everything there is to see. We are nowhere near either condition. Every order of magnitude improvement in resolution has historically opened up entirely new fields of inquiry rather than closing them down–bacteriology, virology, molecular biology, nanotechnology. Each frontier seemed like it might be the last one. None of them were.
AI looks like the same dynamic operating across every domain of human knowledge simultaneously. The flat part of the diminishing marginal returns curve keeps moving outward because understanding the world at a deeper level reveals a world that is deeper than we thought.
My current trails are hilly but not mountainous, so I just use one most of the time.
The lawyer business will be interesting to watch because AI can’t be dragged into court for cross examination as to why certain papers and decisions were made. A human will still need to face judges and politicians. AI can help compile legal precedents and polish briefs but humans have to sign off on them. AI can’t read a room and make the split second decisions as to what to emphasize in legal points and what not. Paralegals will still be needed for the small stuff even if they utilize AI. AI can’t be indemnified, legally deposed, or swear in a court room so help them God. I suspect it will be judges at all levels, local, state, and Federal who will yell....”get me humans or so help me there will be no trials in this court room!”
Yes but AI can’t read the room and make arguments that sway a jury or judge. It will be the human element that retard’s AI uses in court rooms. You can’t make AI swear to tell the truth. Some will try and cross examine and thus the dusty lawbooks will have to be brought out that prove the AI compiled cases were true, not made up, or citing laws within proper context. Judges themselves will balk, holding up cases until; they are assured it’s all humans and not machines.
With problems already being noted with AI systems making up statistics and events to support their arguments, the judges will shut down or limit AI’s use, despite what corporate interests would prefer.
Perhaps Vonnegut living in our time would have noted the conditions today as they are emerging and would have written a book about what you are describing and called it....”A Lead Pipe Cinch” using that starkly laid out irony laden dry narrative Vonnegut was famous for.
Once a trust fund kid before the great FIAT money burn and a lawyer before LAWBRAIN matrix bots replaced him. ...”ah but can he dig a ditch and lay down pipe?” said Zubin Patel, construction site manager.
Excellent!
Disclaimer: Opinions posted on Free Republic are those of the individual posters and do not necessarily represent the opinion of Free Republic or its management. All materials posted herein are protected by copyright law and the exemption for fair use of copyrighted works.