You want one safe career to bet on. The jobs AI can't replace exist, but they ask a lot of your body, your time and your pay.
This in-depth guide covers everything you need to know about ai proof jobs: what jobs are safe from ai, and the cost. Based on verified income data and real-world case studies from our database of 138 side hustle tactics.
No job is fully AI proof, but some are far less exposed than others. The jobs least touched by AI today are physical, local, licensed or care work: roofers, nursing assistants, home health aides, plumbers, dental hygienists, massage therapists. Many of them pay below the national median, are hard on the body or take years of training, and the "safe" list gets shorter each year as warehouse robots, humanoids and AI agents improve.
The question you are really asking
You typed "what jobs are safe from AI" because something shifted. Maybe a manager mentioned "efficiency" in a meeting and looked at your team. Maybe your kid is choosing a university course and you no longer know what to tell them. Maybe you are the one with the degree, the loan and the rent, and you have started to wonder whether you should have learned a trade instead.
So you want a list. A short, calm list of jobs that will still be there in ten years, so you can make one good decision and stop worrying. The honest answer is that the list exists, but it comes with costs nobody mentions in the headline: lower pay in many cases, sore knees and early mornings, and years of training before you earn a full wage. And the list is shrinking.
What the exposure data actually measures
Three large studies try to measure which jobs AI touches. They use different methods, and they point the same way.
Microsoft Research analysed about 200,000 anonymised conversations with Bing Copilot and matched them to the work activities of every US occupation. The paper, first published in July 2025 and revised in December 2025, gives each of 785 occupations an AI applicability score. The scores are public. Interpreters and translators rank first. At the bottom, eight occupations score exactly zero: orderlies, floor sanders, pile driver operators, rail-track equipment operators, foundry mold makers, water treatment plant operators, bridge and lock tenders and dredge operators. Just above them sit motorboat operators, logging equipment operators, maids and housekeeping cleaners, roofers and massage therapists.
The earlier "GPTs are GPTs" paper by researchers at OpenAI and the University of Pennsylvania, first released in March 2023, estimated that around 80% of the US workforce could have at least 10% of their tasks affected by large language models, and about 19% could see at least half their tasks affected. It found that exposure rises with income. It also found that jobs leaning on "science and critical thinking skills" are "strongly negatively associated with exposure", while programming and writing skills are strongly positively associated. Its appendix lists 34 occupations where no task was rated as exposed at all. They include dishwashers, short order cooks, cement masons, stonemasons, power-line installers, motorcycle mechanics, tire changers, roofers' helpers, slaughterers and meat packers, and athletes.
Anthropic, the company behind Claude, publishes its own Economic Index from real usage. Its first report in February 2025 found that "occupations involving a high degree of physical labor" were the least represented, and that the lowest-use jobs at both ends of the pay scale were "generally jobs involving a large degree of manual dexterity, such as shampooers and obstetricians." Its June 2026 report says transportation and material moving, food preparation and serving, and construction and extraction are still under-represented in Claude sessions.
Read together, the three studies agree. If the job needs your hands on a real object in a real place, today's AI barely registers.
The least exposed jobs, with pay and growth
The table puts Microsoft's ranks next to US Bureau of Labor Statistics median pay for 2025 and the official projections for 2025 to 2035, released on August 27, 2026. A rank near 785 means very little overlap with what people successfully do with AI. For context, the median wage for all US workers was $50,980 in May 2025, and total employment is projected to grow 3.5%.
| Occupation | AI applicability rank (of 785) | Median pay 2025 | Projected change 2025 to 2035 | Typical way in |
|---|
| Roofers | 770 | $55,440 | +5% | No formal credential, on-the-job training |
| Massage therapists | 765 | $58,450 | +15% | Postsecondary program, state licence in most states |
| Phlebotomists | 754 | $45,230 | +7% | Postsecondary nondegree award |
| Nursing assistants | 745 | $41,870 | +3% | State-approved program and exam |
| Construction laborers | 740 | $46,680 | +6% | No formal education required |
| Licensed practical nurses | 732 | $64,400 | +3% | About 1 year of training plus licensing exam |
| Physical therapist assistants | 711 | $68,380 | +23% | Associate's degree and licence |
| Dentists | 705 | $176,110 | +6% | Doctoral degree |
| Home health and personal care aides | 689 | $35,800 | +18% | High school diploma |
| Dental hygienists | 667 | $98,100 | +8% | Associate's degree, licence in every state |
| Firefighters | 626 | $59,280 | +4% | Academy training |
| Plumbers, pipefitters and steamfitters | 613 | $63,800 | +7% | Apprenticeship of 4 to 5 years |
| HVAC mechanics and installers | 472 | $61,010 | +11% | Postsecondary nondegree award |
| Electricians | 371 | $63,190 | +9% | Apprenticeship of 4 to 5 years |
Ranks are from Microsoft Research's working-with-ai data. Pay and projections are from each occupation's page in the BLS Occupational Outlook Handbook, linked in the table. Nursing assistant pay is the combined figure for nursing assistants and orderlies.
Electricians sit in the middle of the rankings, at 371, because planning, reading codes and quoting jobs are information work. They are on the list because they are among the most popular "safe" choices, and because the physical half of the job is still out of AI's reach.
The price of safety
Look at the pay column again. Four of the 14 jobs in the table pay below the $50,980 national median: home health aides at $35,800, nursing assistants at $41,870, phlebotomists at $45,230 and construction laborers at $46,680. Further down the exposure list it gets worse. The largest single group of low-exposure workers in the BLS data, hand laborers and material movers, is 6.9 million people with a median pay of $38,220. Janitors and building cleaners earn a median of $36,840, with growth of only 2%.
The jobs that are low-exposure and well paid ask for something else in return. Dentists need a doctoral degree. Electricians and plumbers typically serve an apprenticeship of 4 to 5 years, learning on a trainee wage before they reach the median. Dental hygienists need an associate's degree and a licence in every state.
Then there is the body. BLS says plainly that construction laborers "have one of the highest rates of injuries and illnesses of all occupations." Roofing means heat, ladders and falls. Nursing assistants lift and turn patients through night shifts. These are jobs that many people cannot do at 55, and that a desk worker with a bad back cannot simply switch into.
There is also the matter of place. A plumber's work exists where the pipes are. That protects it from being done by software in another country, and it ties your income to the local economy and to how many hours your body can stand on a job site.
Question to sit with: if the safest jobs on paper are ones you would not want your own child to do for 40 years, what does "safe" actually mean to you?
Care work is growing for a reason that has little to do with AI
The strongest growth in the table comes from care. Home health and personal care aides are projected to grow 18%, physical therapist assistants 23% and massage therapists 15%. BLS says demand for aides will rise as the older population grows, because these services are "in high demand to care for the rising number of older people." Across the economy, private healthcare and social assistance is projected to add over 2.2 million jobs, about 37% of all new jobs through 2035, and healthcare support is the fastest-growing major occupational group at 13.3%.
The World Economic Forum's survey of over 1,000 employers points the same way. Its Future of Jobs Report 2025 expects the largest growth in absolute numbers to come from frontline roles: farmworkers, delivery drivers, construction workers, salespersons and food processing workers, with care roles such as nursing professionals and personal care aides also growing strongly.
The driver here is an ageing population. The work will be there for decades, and the pay in much of it will stay low: the fastest-growing care job in the table also has the lowest pay.
Robots are working through the "safe" list
The exposure studies above measure overlap with chatbots and language models. They say very little about machines with arms and wheels, and that is where the "safe" list is being tested.
Amazon said in mid-2025 that it had deployed its one millionth robot, and reports at the time described the company as having nearly as many robots in its warehouses as human workers. In October 2025 The New York Times reported on internal documents showing Amazon's robotics team hoped to avoid hiring more than 600,000 US workers by 2033 and to automate 75% of its operations, as summarised by the New York Daily News. Amazon said the documents gave an incomplete and misleading picture.
The BLS projections already show the effect in warehouses. Industrial truck and tractor operators, the people who drive forklifts, rank 764 of 785 for AI applicability, almost no exposure to chatbots. Yet their projected growth is only 1% over ten years, with BLS pointing to the "expansion of automated machinery and technologies." Machine feeders and offbearers are flagged on the same handbook pages as tasks that "are likely to continue to be automated."
Factories are further along. The International Federation of Robotics reported in September 2026 that the global stock of industrial robots rose 9% to a record 5 million units in 2025, with more than 600,000 installed that year. In professional service robots, about 7,000 full-size humanoids were sold in 2025.
Humanoids have started doing real shifts. Figure AI says its Figure 02 robots worked an 11-month deployment at BMW's Spartanburg plant, on 10-hour shifts, loading more than 90,000 sheet metal parts and contributing to more than 30,000 BMW X3s. At CES in January 2026, Hyundai and Boston Dynamics showed a product version of Atlas that will be deployed by 2028 at Hyundai's EV plant near Savannah, Georgia to help assemble cars.
Robots are also aiming at the home. 1X opened preorders for its NEO household humanoid in October 2025 at $20,000, or $499 a month, with US deliveries targeted for 2026. That is the job of maids and housekeepers, who rank 773 of 785 on Microsoft's list.
Driving was once on everyone's safe list too. In February 2026 Waymo said it was providing about 400,000 paid robotaxi rides per week across six US cities, and aiming for more than 1 million weekly by the end of the year.
The OECD reached a conclusion in 2023 that surprises people who think manual work is safe. In its Employment Outlook 2023, it found that occupations at the highest risk of automation make up about 27% of jobs, and that they are concentrated in low and middle skilled work, including construction, farming, fishing and forestry.
Question to sit with: if you retrain for a physical job to escape AI, how many of that job's tasks could a robot that costs $20,000 do in ten years?
The safe list keeps getting shorter
The 2023 study listed 34 occupations with zero exposed tasks. The tools have changed a great deal since then. METR, a research group that tests AI agents, reported in March 2025 that the length of tasks frontier AI agents can complete on their own has been "doubling approximately every 7 months" over six years. Agents that can work for hours on their own reach further into the paperwork side of physical jobs: quoting, scheduling, billing, compliance, diagnosis notes. The trade survives. The office around it shrinks.
Even the people building AI talk about physical work as a delay rather than a wall. Geoffrey Hinton, the AI researcher who shared the 2024 Nobel Prize in Physics, told the Diary of a CEO podcast in June 2025, as reported by Business Insider: "I'd say it's going to be a long time before it's as good at physical manipulation." He followed it with "So a good bet would be to be a plumber." In the same interview he said the technology will eventually "get to be better than us at everything."
There is a second squeeze that rarely makes the lists. When AI pushes people out of office work, some of them retrain into trades and care. That increases the supply of workers chasing the same safe jobs, which can hold down the wages of everyone already doing them. That is an inference from how labour markets work, and no official forecast has measured it yet. Think about it before you pay for a course.
Where the evidence is more reassuring
Some of the robotics news is less advanced than the headlines suggest. Figure retired its Figure 02 fleet after the BMW pilot. When the Wall Street Journal tested 1X's NEO, it could not perform any task autonomously, and the chief executive said "much of the work will be done by teleoperators in the beginning." Hyundai's Atlas plant deployment is a 2028 plan. A roof in the rain, a crawl space under an old house or a frightened patient at 3am are still very hard problems for a machine.
The official numbers also show demand growing for most of the jobs in the table, from 3% for nursing assistants to 23% for physical therapist assistants, against 3.5% for all jobs. For the next decade, hands-on, licensed and care work is the least exposed part of the economy. That is the honest good news, with the costs attached.
If you want the other side of the picture, our series covers who is losing work first in jobs AI will replace, the numbers behind the forecasts in how many jobs will AI replace, and two of the jobs people assume are safe in will AI replace nurses and will AI replace doctors.
Question to sit with: are you looking for a job that AI cannot touch, or for an income that does not depend on any single job?
What this means for you this month
The useful step this month is smaller than a career change: find out how exposed your own income is, and whether you have anything that does not depend on one employer.
Write down the five tasks that fill most of your week. Mark each one as information work (writing, checking, summarising, scheduling) or as work that needs your hands, your presence or your licence. The more of your week sits in the first group, the more urgent it is to build a second source of income.
If a trade or care role appeals to you, price it honestly before you commit: training time, the trainee wage, and what your body can do at 50. Reading how to change careers at 40 first will save you some expensive mistakes, and how much money you need before you quit tells you how long your savings can carry a retraining gap.
If you would rather use what you already know, online tutoring pays by the hour for skills that need a person in the room or on the call, and AI labs pay people with degrees to grade model output, covered in our guide to AI training jobs.