The ‘messy middle’ arrives before the apocalypse
The AI jobs debate keeps getting flattened into two loud, lazy camps. One side sees a near-term labor-market wipeout, with whole occupations evaporating before anyone’s time to update a résumé. The other side shrugs and says the panic’s mostly vibes, that companies are still buying software and workers are still collecting paychecks, so what’s the fuss?
Molly Kinder wants neither version.
Her framing’s simpler, and a lot less comforting. First, there’s the labor market as it exists right now, which still looks mostly intact in the aggregate. Then there’s the fully automated future that gets waved around in pitch decks, panel talks and the more breathless corners of tech news. In between sits what she calls the “messy middle,” a long stretch where AI changes tasks inside jobs without flattening the entire labor market all at once.
The panic may arrive in headlines before it shows up in the payroll data.
That middle period’s where a lot of the real work happens, and it’s also where the argument gets politically tricky. If the economy were about to be hit by one giant, obvious wave of displacement, policymakers could treat it like a disaster and respond in emergency mode. But if the change comes job by job, task by task, sector by sector, the damage may be easier to dismiss right up until it becomes hard to ignore.
So far, the evidence for broad labor-market carnage’s limited. That doesn’t mean nothing is happening. It means the machine is still early in its cycle. Firms are testing tools, folding them into workflows and trimming some tasks around the edges. A few people will feel that first. Most won’t, at least not yet. For now, the numbers don’t show a sweeping collapse that’d let anyone say, with a straight face, that the whole white-collar economy has already been upended.
Naturally, that restraint matters. Kinder’s warning isn’t that AI has already blown up work. It’s that partial automation can drag on for years, which makes it easier for everyone to misread the scale of the problem. A company may cut a team here, freeze hiring there and ask the remaining staff to do more with a chatbot nearby. On paper, that can look tidy. In real life, it can hollow out career ladders one rung at a time.
And that’s where the politics get sharper. A slow-moving shift can still be brutal if the losses pile up in a few places: a handful of occupations, specific cities, certain wage bands, certain age groups. The backlash won’t wait for some dramatic all-at-once crisis, if enough people feel the floor moving under them while the broader economy keeps humming. It’ll build quietly, then show up loudly, which is usually how these things go.
For Kinder, that’s the real threat buried inside the current ai policy debate. The “messy middle” may not look like a cinematic jobs apocalypse. It may look smaller than that. It may also prove harder to manage, because it arrives in pieces and keeps asking for a policy response long before everyone agrees there’s a problem worth naming.
Why white-collar workers are first in line
That “messy middle” Kinder talks about doesn’t arrive evenly. Right now, the most exposed workers are the ones Silicon Valley usually talks about as if they were safely above the fray: people whose jobs live on screens, in inboxes and in spreadsheets.
The reason is pretty plain. Large language models are built to handle tasks that are already digital, text-heavy, and repeatable. They read, summarize, draft, classify, search, sort, and rewrite. That maps neatly onto a lot of office work. It maps much less neatly onto a person fixing a leaky pipe, trimming hair, loading shelves, or carrying plates through a crowded dining room. A review from the International Labour Organization on generative AI and work organization makes the same basic point: the earliest exposure shows up most clearly in tasks that can be copied into software, not in jobs that depend on hands, bodies, and a physical setting.
The safest jobs right now are often the ones that still require a hand on a tool or a person in the room.
That’s why the first pressure points sit in law, finance, consulting, sales and the clerical back office. Legal work’s plenty of document review, contract comparison and research that software can now chew through faster than a junior associate with three coffees and a deadline. Finance’s reporting, analysis and routine client communication. Consulting firms are already watching AI draft slides and pull together first-pass recommendations. Sales teams use it for outreach, call notes and follow-up. Back-office clerical jobs, which often live on forms, records, and repetitive admin, are even more exposed because so much of the work is already standardized.
White-collar work also tends to be easier to break into pieces. A model doesn’t need to replace an entire paralegal or analyst to create trouble. It only has to nibble away at the billing hours, the entry-level tasks, and the routine assignments that used to keep junior staff busy long enough to learn the job. That’s where the anxiety starts, and it’s why the story of AI job loss so far has been less about dramatic layoff headlines than about quiet reordering inside teams.
By contrast, blue-collar, service and workplace-based jobs still have a lot of insulation. Restaurant staff need to move through real space, handle unpredictable customer behavior and deal with a kitchen that doesn’t care about your prompt-writing skills. Salon workers need dexterity, trust and a chair-side sense of what a person actually wants. Repair workers need to look at a broken thing, touch it, test it and improvise when the problem turns out to be weirder than the manual promised. AI chat tools don’t do much for that kind of labor today. They can help with scheduling or writing a text to a customer, sure. They can’t tighten the bolt.
That does not mean physical work is safe forever. Robotics will matter, and maybe sooner than the skeptics want to admit. But the pace is different. Robots have to deal with the real world, where tables wobble, floors get slippery, and every doorway seems designed by an enemy of efficiency. Knowledge work, by contrast, is already sitting inside a computer. That is why the adoption curve is steeper there, and why policy makers are paying so much attention to office jobs in the latest America’s AI Action Plan, which treats workforce adjustment as a live issue rather than a hypothetical.
Kinder’s line on this is sharp enough to survive the policy jargon. If you can do your job shut in a room with a computer, you’re eventually at risk. That sounds a little rude at first, but it’s hard to argue with. The first wave isn’t coming for every worker at once. It’s coming for the people whose labor can be turned into text, codes and tidy outputs before lunch. And that, for all the jokes about robots, is a very human kind of problem.
What gets broken when the best jobs are the ones at risk
For a long stretch of American work history, automation mostly pressed on the jobs people were trying to get out of. Farm labor shrank. Factory work thinned out. The old promise of office life was different: if you could land a desk, a degree and a decent title, the machine would usually help, not replace, you. Computers handled the repetitive stuff. They sorted invoices, crunched numbers, stored files and made it possible to do more work with fewer mistakes. The human part of the job still sat in the middle.
That bargain is what makes the current wave feel so odd. Large language models don’t just trim the boring edges of white-collar work. In a growing number of cases, they can draft the memo, summarize the case, build the presentation, write the first pass, and produce something a manager can approve with very little cleanup. The balance starts to tilt. The worker is no longer the person using the tool to get to the real value. The tool begins to look like the thing producing the value, while the worker becomes a reviewer, fixer, or, in some settings, a luxury. The International Labour Organization’s occupational exposure index for generative AI maps that risk across occupations, and the pattern is hard to miss: office jobs are sitting much closer to the blast radius than people expected.
The unnerving part is not that AI is coming for the jobs nobody wanted. It’s that it may arrive first for the jobs people spent years trying to earn.
That is why the anxiety feels different from earlier automation scares. Lawyers, doctors, professors, market researchers, analysts. These are the careers parents still point to when they say, “Stay in school and you’ll be fine.” They are also the jobs that many students have been trained to chase through years of debt, internships, unpaid labor, and the ritual humiliation of entry-level hiring. If parts of those roles get commoditized, the damage won’t stop at the office. It spills into mortgage applications, apartment searches, wedding plans, child care budgets, and the basic math of staying in the middle class.
The weak spot here isn’t just pay. It’s trust. For a generation or two, education’s been treated as a fairly reliable trade. Put in the tuition, survive the exams, collect the credential and the labor market would usually cough up a stable salary. That story was never perfect, but it held often enough to shape family expectations. Now that bargain looks shakier. A professor who can be partly replaced by software, an analyst whose first draft comes from a model, a junior lawyer whose research memo takes minutes instead of hours, a doctor whose paperwork’s swallowed by an AI system, all of that chips away at the old confidence that white-collar jobs were the safe lane.
There’s a timing problem too. The people most exposed are often young enough to have borrowed heavily for the promise of entry. They’re also old enough to be planning homes, partners, and kids around the idea that their first real job will turn into a career, not a temporary perch. When that expectation slips, the mood changes fast. You can hear it in students who did what they were told and are now watching the payoff wobble before they’ve even started paying down the loans. And you can hear it in early-career knowledge workers who can already see a model do in seconds what took them an afternoon.
And none of this requires a robot arm in a warehouse. The pressure is already inside the laptop class, where knowledge workers have spent years assuming that their value came from judgment, taste, and expertise. That may still be true. For now, though, the software is getting good enough to take over more of the routine thinking, and the line between “assistant” and “replacement” keeps getting fuzzier. A faster frontier model, including China’s latest entrant, only makes that race feel less theoretical and more like a moving target.
The result is a very specific kind of dread. Not the old fear that machines would take the jobs humans never wanted. This is the fear that the ladder itself’s being pulled up while people are still climbing it.
Why Kinder rejects a universal check
After the white-collar risk picture gets clear, the policy argument stops being abstract very quickly. Molly Kinder’s view’s basically this: if generative AI starts knocking people out of salaries they spent years building, the answer shouldn’t be a shrug and a monthly deposit that treats work like a hobby people can pick up or drop whenever. A universal basic income may sound tidy in a panel discussion. In the labor market, it gets messy in a hurry.
Kinder’s objection’s practical, not sentimental. If a laid-off software engineer, analyst, or junior lawyer gets the same flat check no matter what happens next, why should anyone keep signing up for the unpleasant but necessary jobs that keep hospitals staffed, apartments repaired and streets safer? That question gets sharper, not softer, when the disruption comes from job displacement in the white-collar ranks. A paycheck can cushion a fall. It can’t, on its own, rebuild the ladder that was kicked away.
The point is not to turn labor into a side quest for people who lost their old jobs.
That’s why she favors targeted tools instead of a universal payout. The goal, in her telling, is to manage the transition, not to surrender to it. A broad cash transfer can reduce hardship, sure. It can also freeze in place a world where the people with the best credentials are told to sit still and wait, while the jobs that still need human hands get harder to fill. That’s a strange bargain to make in a labor market already being changed by generative AI.
One of Kinder’s sharper ideas is a workforce reinvestment fund. The basic premise’s simple enough to survive a congressional hearing, which is saying something. Companies that trim entry-level hiring would pay into a pool that helps finance white-collar apprenticeships and other early-career pathways. The logic’s part penalty, part repair. They should help pay for a new rung, if firms save money by cutting the first rung of the career ladder. No one gets to cheerfully automate the receptionist, the research assistant and the junior associate, then act surprised when graduates start sending very pointed emails.
That fund matters because the first jobs lost in an AI transition are often the jobs that teach people how to work. The damage spreads beyond the workers who were cut this quarter, if those disappear. The entire pipeline gets thinner. Young workers lose the chance to build experience, and employers later complain that nobody has experience. A bit circular, that.
From there, Kinder also wants wage insurance for older workers who are forced to switch roles later in life. That’s a different problem. A 24-year-old who loses an entry-level role can still spend a few years rebuilding. A 52-year-old mortgage holder with teenagers in school has a tighter margin for error and a lot less appetite for a fresh start in a field they never planned to enter. Wage insurance wouldn’t make the switch painless. But it could soften the drop when someone has to move from one profession to another and take a pay cut to do it. That seems a lot more humane than pretending everyone can simply upskill their way into serenity.
There’s also a stronger political edge to her thinking. If the private labor market stops producing good jobs on its own, she’s arguing for a public effort to create them. Call it industrial policy for knowledge workers, though the label’s uglier than the idea. The principle’s straightforward: if AI keeps eating the entry-level tasks that used to train people into stable careers, then government should help build new ones. That might mean subsidized apprenticeships, employer incentives, or direct support for roles that the market is no longer bothering to grow on its own.
That approach fits the scale of the problem much better than a universal check does. The ILO-NASK index on generative AI and jobs puts a number on the churn by saying roughly one in four jobs could be transformed. Transformation is not the same as elimination, and that distinction matters. It means policymakers still have room to slow the damage, spread out the pain, and keep people attached to work.
For Kinder, that’s the whole point. Don’t pay people to disappear into the background. Keep the labor market functioning long enough for it to adjust, and when it fails to do that, build something better before the bill arrives. The next question’s who does that work, and what kind of institution can actually carry it.
AI policy’s next test is institutional
Kinder’s next move says a lot about where this debate’s headed. After roughly three years at Brookings, she’s leaving to launch a new group focused on AI-transition problems, which is a very Washington way of admitting that commentary alone won’t fix a labor market that starts fraying in the wrong places. The pitch now is less about making predictions and more about building something that can actually absorb the shock.
That matters because the central question’s moved past the familiar headline fight over whether AI will create jobs or kill them. It probably will do both, though not in neat little ledger entries. The real issue’s whether policy can handle concentrated white-collar losses without making inequality worse. The damage won’t stay neatly contained inside one company or one quarter’s earnings call, if the first jobs thinned out by AI are the entry-level office roles that used to train the next generation. It spills into paychecks, career ladders and who gets to stay inside the middle class.
The political damage will come from the people who thought they had time.
That’s why Kinder keeps coming back to pace. This doesn’t look like a movie scene where a whole sector disappears before lunch. It looks slower, and that’s almost trickier. Firms may trim junior hiring first. Then they may ask a smaller group to do more drafting, more sorting, more routine analysis. Then the pipeline gets narrower. A few years later, employers complain they can’t find experienced workers, while younger workers complain they were never given a real chance to become experienced in the first place. Charming system.
So the challenge for workforce policy isn’t just to soften the blow after layoffs show up in the data. It’s to stay ahead of them. That means retraining that’s tied to real openings, not inspirational slide decks. It means wage insurance for people who need to switch roles later in life and can’t simply start over with a smile and a badge reel. It means creating jobs where the market stops producing them, which is where public policy tends to get much less glamorous and much more useful.
The timing matters too. If AI disruption lands over years, then politics will also move over years, which gives leaders a choice most of them will probably try to avoid: act before the pain’s obvious, or wait until every office worker with a mortgage and a student loan’s figured out what automation feels like up close. By then, the backlash writes itself.
Kinder’s new organization seems built around that problem, not the abstract version of it. The work ahead’s institutional in the plainest sense. Someone has to design the rules, the training, the transition aid, and the job creation efforts before the losses become a headline people can’t ignore. In the end, AI policy won’t be judged by the demos or the speeches. It’ll be judged by what happens to the people who were supposed to be safe.
