I keep a scrap of notebook paper from a conversation I had last spring with a warehouse supervisor outside Columbus. He was not panicking about robots. He was annoyed. His best picker had just left for a dental-assistant program, and two more were quietly studying for commercial driving tests on their phones during breaks. “The work is still here,” he told me. “The people who want this particular work are not.” That scrap came back to me when a widely cited workforce study put a number on the same feeling: roughly 11 million US workers may need to switch occupations by 2035. Not vanish. Switch. About 7 percent of today’s workforce. The distinction sounds small until you sit with it.
Headline culture prefers the cleaner story. Tens of millions erased. A billion deaths whispered in the same breath as software updates. I have found that the cleaner story is usually the less useful one. The more interesting claim is messier. Job openings can be plentiful and still leave millions stuck, because the openings demand different skills, different credentials, different cities, or different pay structures than the jobs people already hold. That is not science fiction. It is a matching problem with a calendar attached.
What The 11 Million Figure Actually Describes
Start with the scale, because scale is where the panic usually begins and where it usually goes wrong. Eleven million people is a large room. It is also, on the study’s own framing, a minority of the labor force. Seven percent is not nothing. It is also not the end of paid work. The research treats the coming decade as a transitional period concentrated among people in lower-skill roles, not as a cliff that swallows every occupation at once.
The sharper line is about money. Six out of seven of the workers who would need to move could face income loss in the process. Read that twice. The issue is not only whether a new task exists. It is whether the new task pays like the old one, sits in a reachable place, and accepts the credentials a person already has. A forklift operator who becomes a home-health aide has switched careers. If the hourly rate drops and the schedule fragments, the switch is a demotion wearing a new name tag.
Job opportunities can be abundant and yet leave millions of workers without work if those positions require different skills, credentials, locations or pay structures than current jobs.
Workforce research summary, paraphrased
I have sat through enough labor briefings to know how easily that sentence gets flattened into “AI takes the jobs.” It does not say that. It says the market can clear on paper and still fail particular people. That is an older problem than chatbots. It is also the problem worth planning for.
A Transition, Not A Vanishing Act
Perhaps the most useful mental model is a highway interchange, not a sinkhole. Traffic keeps moving. Some lanes close. Some exits require a different vehicle. Drivers who refuse the exit sit in a queue that looks, from a distance, like unemployment. Up close it is a skills queue, a geography queue, or a pay queue.
Recent labor analysis frames the disruption this way on purpose. Demand for some routine cognitive tasks falls. Demand for hands-on care, skilled trades, and coordination work holds or rises. The net can look calm while the gross flows are violent. People do not experience net. They experience the Tuesday when their shift pattern changes and the certification they need costs four months of rent.
In my experience, the people who handle this best are rarely the ones quoting productivity statistics. They are the ones who treat a credential like a spare tire. Annoying to buy. Obvious once you need it.
Who Sits Closest To The Exit Ramp
Exposure is not evenly sprinkled. Office support, routine customer handling, basic data entry, parts of food service, and some logistics coordination show up again and again in automation studies. So do slices of back-office finance and insurance processing. The pattern is boring, which is why it is easy to ignore. If a task is repetitive, text-heavy, and already half-digitized, software is going to nibble it.
That does not mean every clerk disappears. It means the clerk who only moves fields between screens is in a thinner market than the clerk who can explain an exception to an angry customer and then fix the system that created the exception. Same job title. Different durability.
- Roles built on predictable screen work face the fastest redesign.
- Roles built on physical presence, judgment under mess, or licensed care hold up better.
- Hybrid roles, part routine and part exception-handling, split in the middle.
- Geography decides whether a “better” role is actually reachable.
- Pay structure decides whether the move feels like progress.
Age matters too, and not in the cartoon way. A 28-year-old can absorb a credential detour. A 54-year-old with a mortgage and a parent in assisted living cannot treat two years of school as a lifestyle choice. The study’s income-loss warning lands hardest on that second group. Career switch sounds entrepreneurial until the calendar is short.
Why Six Out Of Seven Is The Number That Should Stick
Headcount gets the headline. Pay gets the bruise. If six of every seven affected workers risk earning less after the move, the social story is not “new opportunities bloom.” It is “opportunities bloom at a discount.” That discount can be temporary. It can also harden into a lower track, especially if the new occupation has weaker bargaining power or more variable hours.
Think about a claims processor who shifts into retail supervision, or a junior paralegal who lands in customer success at a software firm. Both switched. Neither is guaranteed the old salary, the old benefits, or the old predictability. Benefits are the quiet half of compensation. Lose the health plan and the hourly rate can rise while the household falls behind.
I would rather we argue about that discount in public than pretend the only question is whether a job posting exists. Postings are cheap. Matching is expensive.
We Have Done This Before, Just Slower
Nothing about a labor-market scramble is new. Tractors emptied fields. Assembly lines, then automated assembly lines, rearranged factories. Office computers shrank typing pools. Online retail rearranged store floors and warehouses in the same decade. Each wave lifted output. Each wave also stranded people whose skills were tuned to the previous machine.
The comforting version of that history says living standards rose, so relax. The accurate version says living standards rose on average, after a lag, and not for every town that lost the mill. Averages are a poor blanket. They do not cover the county where the new jobs require a license the community college does not offer.
What feels different this time is the speed of the interface, not the existence of change. Software can be updated on a Thursday. A person’s mortgage cannot. That gap between deployment speed and human retraining speed is the whole policy argument, whether or not anyone wants to admit it.
The Doom Narrative Has A Business Model
Here is the part I distrust. Extreme forecasts travel well. A claim that software will erase tens of millions of jobs, or that advanced systems could somehow be tied to mass death, gets shared. Milder claims about occupational mobility get skimmed. The extreme version also flatters a regulatory instinct: if the tool is apocalyptic, the people who already build it should help write the rules. Moats love fear.
I am not arguing that risk is fake. I am arguing that inflated risk is a product. Climate debates taught a generation of communicators that catastrophic framing moves money and statutes faster than a table of trade-offs. The same muscle is being used on workplace software. A sober reading of the 11 million figure is less useful to that muscle. It suggests adaptation, credentials, and local labor markets. Those are unglamorous. They also happen to be the levers that work.
Perhaps the fairest test is this. If a prediction cannot distinguish between a cashier and a nurse, it is not a labor forecast. It is a mood.
Where Demand Is Likely To Pool
Look at the other side of the ledger, because a career switch only makes sense if something is hiring. Care work is not optional in an aging country. Skilled trades keep showing shortages that software does not patch. Installation, maintenance, and field service sit at the awkward intersection of hands and diagnostics. Health support roles, logistics exceptions, and specialized construction keep appearing in employer surveys even as clerical headcount plans shrink.
None of that is a promise. A shortage on a chart is not a job offer in your zip code. Still, the direction of demand is clearer than the doom reels suggest. The economy is not running out of tasks. It is running out of tasks that look like 2014 back-office work.
| Pressure Point | What Tends To Shrink | What Tends To Hold |
| Routine screen tasks | Data entry, basic processing | Exception handling, client judgment |
| Physical presence | Simple repetitive picking | Maintenance, licensed care, field work |
| Pay structure | Hourly roles with thin benefits | Credentialed roles with stable schedules |
| Location | Single-employer towns | Regions with mixed employers |
| Time horizon | Roles already half-automated | Roles needing trust or a license |
Tables like that are blunt. Real workplaces blur the columns. A hospital still needs schedulers. A warehouse still needs people who can override a bad pick path. The point is directional. If your day is mostly copying, sorting, and confirming, assume the day will be redesigned. If your day is mostly deciding under incomplete information while standing next to another human, you have more time.
Skills Are The Toll, Not The Slogan
Everyone says “learn new skills” the way people say “drink more water.” Fine. Incomplete. The skills that transfer are narrower than motivational posters admit. Communication under stress. Basic data literacy, meaning you can tell when a dashboard is lying. A license that an employer cannot fake. The ability to document a process so someone else can run it. Those travel. A vague promise to “get into tech” often does not.
Credential friction is the part households feel and consultants underweight. A six-week badge is not the same as a state license. Employers know the difference even when marketing pages pretend otherwise. I have watched people burn savings on a certificate that local hiring managers do not recognize. That is not a personal failure. It is a market with bad labels.
- Name the task you actually do, not the title on the badge.
- Ask which neighboring occupation already hires people with that task.
- Check whether a license, not a logo, gates the pay.
- Price the training in rent-months, not in course hours.
- Line up one employer conversation before you quit.
That sequence is unromantic. It is also how the successful switches I have seen actually happen. The unsuccessful ones start with a headline and a credit card.
Geography Is A Skill You Cannot Download
Remote work softened location for a slice of professional jobs. It did not abolish it. Care, trades, logistics, and most health support still happen in a place. If the growing occupations cluster in metros you cannot afford, the national vacancy rate is a rumor. Households make this calculation with school districts and parents nearby, not with a national average wage.
A career switch that requires a move is a different product from a career switch down the road. The first one prices in housing, a spouse’s job, and the loss of informal childcare. Studies that count occupational mobility without counting moving costs will always look cleaner than lived experience. I would haircut any rosy projection by the share of workers who cannot leave their county. That share is large.
Employers Are Not Neutral Bystanders
Firms adopt tools to cut unit cost. That is not a moral surprise. What is optional is how they handle the people attached to the old unit cost. Some redesign roles and train inward. Some post a new req and let attrition do the quiet work. The second path is cheaper on a quarterly slide and more expensive for the town.
Internal mobility is the underused valve. A company that already trusts an employee’s judgment can teach a new interface faster than a stranger can learn the company’s exceptions. Yet internal postings often demand the new skill on day one, which freezes the very people the firm claims to value. If you run a team, that contradiction is yours to fix. If you are the employee, ask early whether a bridge role exists. Silence is an answer.
The economy adapts. Households adapt on a slower clock, and the gap between those clocks is where income gets lost.
Productivity gains are real when tools remove drudgery. They do not automatically land in the paycheck of the person whose drudgery was removed. Bargaining, scarcity, and ownership decide the split. That is an old argument wearing a new interface.
What A Sensible Household Plan Looks Like
You do not need a ten-year forecast to act. You need a twelve-month one. Which part of your week could a competent intern with a good tool already do? Which part requires your name on a license, your presence in a room, or your memory of how this particular client behaves? Protect the second. Shrink your dependence on the first.
Cash is part of the plan, and I say that without the usual lecture. A career switch with three months of expenses feels like a choice. The same switch with three days of expenses feels like a threat. Income risk is easier to absorb when the household is not already at the edge. That is not profound. It is why the six-out-of-seven warning matters more than the headline count.
A plain twelve-month check: Task map: what is routine vs what is judgment Credential map: what local employers actually require Cash map: months of expenses if pay dips Network map: two people already in the target role Geography map: can the new work be done here
None of those lines require a guru. They require an honest afternoon. Most people skip the afternoon and then feel ambushed by a reorganization email. The email was never the start of the story.
Schools, Licenses, And The Slow Machinery
Community colleges and trade programs are the unfashionable heroes of this transition, when they are funded and when employers actually sit on the advisory board. A program designed in a vacuum produces graduates the local hospital will not hire. A program designed with the hospital’s staffing office produces people who start on a Monday. The difference is meetings, not slogans.
Licensing boards can help or clog. Some gates protect patients and clients. Some gates protect incumbents. If a state makes it irrationally hard to move a credential across a border, it is taxing mobility at the exact moment mobility is the safety valve. I am not romantic about deregulation. I am impatient with rules that do not track the risk they claim to manage.
Apprenticeships deserve more oxygen than another round of generic digital badges. Paid learning attached to a real employer solves the cash problem and the credibility problem at once. They are slower to announce than a partnership press release. They also place people.
The Timeline To 2035 Is Shorter Than It Looks
Nine years sounds spacious until you count school calendars, lease cycles, and the age of the people most exposed. A worker who is 48 today is 57 in 2035. That is not “plenty of time to reinvent.” That is one serious retraining window, maybe two, before retirement math takes over. Planning as if the change arrives on the last day of 2034 is how households get late.
Deployment will be uneven. Some firms will automate a process this year and hire around it. Others will pilot, stall, and pilot again. The aggregate curve can look smooth while individual departments lurch. Smooth curves are a gift to forecasters and a poor guide to your manager’s budget.
So treat 2035 as a horizon, not a start gun. The useful moves are the ones that still help if the forecast is early, late, or half wrong. A recognized credential, a cash buffer, and a skill that shows up in more than one industry survive a bad forecast. A bet on a single job title does not.
Income Loss Is A Design Choice As Much As A Forecast
The finding that most affected workers could earn less is not a law of physics. It reflects how new roles are priced, how benefits attach, and how long it takes to become productive in the next occupation. Employers who slot career-switchers in at the bottom of a scale, regardless of adjacent experience, manufacture the loss. Employers who credit adjacent experience shrink it.
Public programs can blunt the dip with wage insurance, portable benefits, or paid training. They can also miss, if the money arrives after the rent is late. Timing is the whole design. A grant that takes nine months to approve is a press release. A bridge that starts with the layoff notice is a policy.
I do not think every dip should be socialized. I do think pretending the dip will not happen, after research flags it for six in seven people, is a choice to be surprised later. Surprise is expensive.
What Not To Believe
A few claims are circulating that do not survive contact with a staffing office. One: every knowledge job dissolves at once. Hiring managers are still begging for people who can own an outcome. Two: trades are a universal escape hatch. They are a strong hedge, not a door with no line. Three: a single tool certification replaces domain knowledge. It does not. Four: national job growth means your county is fine. It might not be.
There is also the mirror-image claim, that nothing meaningful is changing and the whole discussion is marketing. That one fails too. Task mix inside jobs is already shifting. The colleague who “just uses the tool for the first draft” is describing a real reallocation of time. Reallocation is not extinction. It is still change, and change still has a cost.
A Clearer Way To Talk About Risk
Risk management, in a household or a firm, starts by separating severity from probability. The probability that some occupational switching accelerates is high. The probability that paid work itself disappears is low. The severity of a pay cut for a specific household can be high even when the macro story is mild. Mixing those layers is how we get both complacency and panic from the same spreadsheet.
If you invest, the labor story is a demand story as much as a cost story. Firms that sell tools will talk about replacement. Firms that sell into care, infrastructure, and maintenance will talk about shortage. Both can be right in their lane. A portfolio that assumes only one lane is making a mood trade.
If you manage people, the practical question is which tasks you will stop paying humans to do, and what you will pay them to do instead. Leaving that second half blank is how you create the 11 million from the inside.
Useful question: What must still be true, in person, when the tool is finished?
Weak question: Will robots take everything?I prefer the first question in meetings. It forces a task list. Task lists can be staffed. Apocalypses cannot.
Stories From The Middle Of The Shift
The Columbus supervisor is not a statistic, but he rhymes with one. His pickers were not replaced by a silent fleet overnight. They were comparing their hourly rate with programs that promised a license and a slightly sturdier schedule. Some will earn less for a year. Some will earn more by year three. The study’s six-out-of-seven warning is a statement about the average path, not a verdict on every person who moves.
Another pattern I keep hearing from clinic managers: they cannot fill aide roles, and they cannot interest experienced office staff in the physical work, even with a signing bonus. The jobs are not missing. The translation between “I have done admin” and “I will do nights on a floor” is missing. That translation is cultural as much as financial. Career switch language skips the identity part. People do not only sell hours. They sell a picture of themselves.
Identity is why some switches stall even when the math works. A title can be a smaller paycheck with a larger sense of competence, or the reverse. Households negotiate that in kitchens, not in white papers. Any plan that ignores the kitchen negotiation will overestimate mobility.
How To Read The Next Wave Of Headlines
When the next alarming number lands, ask four plain questions. Is it gross displacement or net employment? Does it separate pay from headcount? Does it say who, or only how many? Does the speaker sell regulation, software, or attention? Those questions will not make you popular in a group chat. They will keep you from rearranging your life around a trailer.
A figure like 11 million is a planning input. It says a noticeable minority may need a different occupation, and most of that minority may earn less at first. It does not say the labor market closes. It does not say your specific role ends on a known date. Used well, it is a nudge toward credentials, cash, and a clearer task map. Used as horror, it is just another scroll.
I keep the notebook scrap because it is smaller than the headlines and more accurate. The work was still there. The people were already choosing different work. That is the transition, in one parking lot. The rest is whether we make the toll cheaper than it has to be.
Practical Moves If You Are In The Exposed Slice
If your week is heavy on routine processing, do not wait for a corporate announcement to draw the map. Talk to two people who left a similar role in the last three years and ask what actually transferred. Look at local postings not for inspiration but for required licenses. Price one credential. Build a small buffer before you need it. Tell your manager you want the exception work, not only the queue. Managers remember the person who asks.
If you are already in care, trades, or field service, the risk is different. It is workload, not disappearance. Protect your license, your body, and your schedule leverage. Shortage can raise pay. It can also burn people out and push them back into the very roles software is thinning. A durable career is not only in demand. It is survivable on a Thursday.
Couples should run the numbers together. A career switch that cuts one income for eighteen months is a household decision, especially with children or shared debt. The romantic version, “follow your pivot,” skips the spreadsheet. The spreadsheet is the kindness.
What Firms Can Do Without A Press Conference
Publish the tasks you expect to automate in the next two years, in plain language, to the people who do them. Offer a bridge role with a training schedule and a pay floor. Credit adjacent experience when you hire internally. Sit on a local program’s advisory board so the curriculum matches your reqs. None of that is revolutionary. All of it reduces the share of switches that become income losses.
Firms that treat the transition as purely a procurement decision will get the cost savings and the turnover. Firms that treat it as a staffing design will keep judgment they already paid to develop. I know which one I would want owning a process that still fails in interesting ways.
The Bottom Line Worth Keeping
About 11 million US workers may need a different occupation by 2035. That is roughly 7 percent of the current workforce, concentrated where tasks are routine and easier to redesign. Six out of seven of those workers could earn less along the way, because skills, credentials, location, and pay structures do not line up by magic. The country has absorbed technological shifts before. Output rose. Some people did not rise with it on the same schedule.
Apocalyptic totals and celebrity warnings about civilizational body counts are a poor guide to that schedule. They flatten a matching problem into a monster story. The matching problem is enough. It is also solvable in pieces: clearer task maps, credentials employers recognize, bridge pay, and a refusal to treat a national vacancy rate as a personal promise.
If you remember one line from the research, make it the awkward one. Jobs can be abundant and still miss the people who need them, when the new jobs ask for a different skill, a different badge, a different town, or a different wage. That is the career switch hiding inside the headline. It is slower than doom, and harder to ignore once you have said it out loud.
