Short answer: the labour market is growing more slowly than it did, and the fastest-growing occupations are concentrated in healthcare, energy and data rather than being wiped out by automation. The US Bureau of Labor Statistics published its 2025-35 projections on 27 August 2026, showing the economy adding 5.9 million jobs over the decade, with total employment rising from 170.3 million to 176.2 million. That is 3.5 percent growth, well below the 10.9 percent recorded over 2015-25. Alongside those projections BLS introduced AI exposure categories for every detailed occupation, which is the first official attempt to quantify this question rather than estimate it.
The Number Everyone Misreads
A 3.5 percent decade is slow by recent standards, and it is not evidence of AI destroying employment. Slower labour force growth, demographic change and sector shifts account for most of it.
What the projections actually show is redistribution. Some sectors are expanding sharply while others contract, and the pattern is more specific than “AI is taking jobs”.
- Utilities is the fastest growing major sector at 9.8 percent, though from a small base, adding 58,800 jobs. BLS attributes nearly all of it to electric power generation, transmission and distribution, driven partly by rising electricity demand including AI power demands.
- Healthcare and social assistance adds the most jobs of any sector, over 2.2 million, about 37 percent of all new jobs projected through 2035, driven by an ageing population and rising chronic disease.
- Professional, scientific and technical services is the third fastest growing at 8.6 percent, adding 926,700 jobs, with demand for AI-based systems, research and development and associated consulting services named as a driver.
- Federal government declines 3.4 percent and retail trade declines slightly, losing about 27,500 jobs as e-commerce continues to limit employment in retail outlets.
Read those together and a clear story emerges: AI is creating employment in the infrastructure and services that build and power it, while the displacement is showing up in retail and administrative work rather than in the professions the coverage focuses on.
The Fastest Growing Occupations
BLS publishes a ranked table, and it is the most useful single artefact for anyone making a career decision this year.
| Occupation | 2025 employment | Projected change to 2035 | 2025 median wage |
|---|---|---|---|
| Nurse practitioners | 336,300 | +41.0% | $132,300 |
| Solar photovoltaic installers | 31,100 | +36.5% | $53,140 |
| Data scientists | 275,600 | +34.6% | $120,230 |
| Wind turbine service technicians | 11,800 | +29.5% | $64,120 |
| Medical and health services managers | 640,400 | +24.2% | $123,860 |
| Computer and information research scientists | 38,600 | +21.8% | $140,300 |
| Information security analysts | 192,900 | +21.0% | $129,180 |
| Home health and personal care aides | 4,677,100 | +18.1% | $35,800 |
| Logisticians | 255,100 | +17.6% | $82,320 |
Source: BLS fastest growing occupations table, 2025 and projected 2035.
Three observations that matter more than the ranking itself.
Data scientists at 34.6 percent is the biggest correction. The occupation most often described as saturated or dying is the third fastest growing in the entire economy, adding 95,400 positions. Whatever is happening to entry-level hiring in that field, the ten-year demand picture is not ambiguous.
Percentage growth and job volume are different questions. Wind turbine technicians grow 29.5 percent and add 3,500 jobs. Home health aides grow 18.1 percent and add 847,300. If you are choosing a career, the percentage tells you about competition and the absolute number tells you about availability.
Wages vary enormously within the growth list. Several of the fastest-growing occupations pay well below the $50,980 all-occupation median. Growth is not the same as opportunity.
What the New AI Exposure Data Actually Says
This is the genuinely new element in the 2026 release, and it is widely misdescribed.
BLS combined five external datasets to sort every detailed occupation into four AI exposure categories: Low, Moderate, High and Very high. Three sources are theoretical measures scoring how well AI capabilities match an occupation’s work. Two are observed-evidence measures built from real usage: one derived from Claude conversations and API traffic, the other from Microsoft Copilot data, both mapped onto occupational tasks.
The methodology is more careful than the coverage suggests. BLS calculated percentile ranks for each of the five measures, normalised them to a common scale, took the median across the three theoretical and the two observed sources separately, then used a clustering algorithm to produce the four categories. Across 4,155 occupation-source combinations covering 831 occupations, 3,944 were observed and 211 imputed.
And then the sentence that should appear in every article on this subject: exposure “does not imply job loss, productivity gains, automation probability, or wage effects.”
BLS is explicit that this product does not measure AI-related employment impacts. It measures how occupations compare with each other on task overlap, to help with career planning. Treating a High exposure rating as a prediction of job loss is a misuse of the data, and it is the most common error in coverage of this release.
How to use exposure ratings properly
As a comparison, not a forecast. The categories are relative. An occupation rated Very high has more task overlap than most others, not a probability attached to it.
Alongside the employment projection for the same occupation. Where the two disagree, that gap is informative. Data scientists show substantial task overlap and 34.6 percent projected growth. Both are true, and the combination tells you the work will change while the demand grows.
As a prompt to look at your own tasks. Occupation-level data cannot tell you about your specific role. Two people with the same job title can have completely different exposure depending on what they actually do all week.
With the theoretical and observed measures distinguished. Theoretical exposure says a capability could apply. Observed measures say usage has been recorded against those tasks. They answer different questions.
Job Displacement, Job Creation and the Skills in Between
The debate about employment trends in the age of AI technology tends to collapse into two positions, and the projections support neither cleanly.
Job displacement is real and narrow. Retail trade loses about 27,500 jobs and federal government declines 3.4 percent. Within occupations, the declines cluster in administrative support, cashiering, claims processing and similar work. These are real losses for real people, and they are concentrated rather than economy-wide.
Job creation is real and physical. The largest absolute creation is in healthcare and social assistance, over 2.2 million positions. The fastest percentage growth includes solar installers and wind turbine technicians. Neither of those is what most coverage of AI technology and the future of work leads with, and both are direct consequences of energy and demographic trends that AI is accelerating rather than causing.
Skills demand is where the two meet. New job roles are appearing faster than occupational classification can track them, which is why they are invisible in the tables. The demand shows up instead as professional and technical services growing 8.6 percent, and as employers asking for applied AI capability inside existing roles rather than hiring for it separately.
Reskilling, training programs and geographic disparities
Three practical factors decide how the transition actually goes for an individual, and none appears in a projection table.
Reskilling works better in adjacent moves. Someone moving from claims processing into claims investigation is using most of what they already know. Someone moving from claims processing into software engineering is not. The successful transitions in every previous wave of technological change were short lateral steps, repeated, rather than single dramatic leaps.
Training programs vary enormously in what they actually deliver. The useful test is whether a program ends with something you can show: a project, a portfolio piece, a certification an employer recognises. Programs that end with completion alone leave you competing on the same evidence as everyone else.
Geographic disparities are widening. Data centre construction, energy infrastructure and healthcare demand are not evenly distributed. Utilities growth concentrates where generation capacity is being built. Healthcare demand follows population age structure. For anyone whose local market is dominated by a shrinking sector, mobility is a bigger lever than reskilling.
There is also a psychological impact worth naming honestly, because it affects decisions. Constant coverage predicting the end of your profession produces anxiety that pushes people either into paralysis or into expensive, poorly chosen retraining. The official data is a useful antidote: it is specific, it is dull, and it mostly does not say what the coverage says it does.
What This Means for Individual Decisions
The projections are ten-year national figures. Translating them into a personal decision requires a few steps.
- Find your occupation’s own projection. Sector-level and top-line figures hide enormous variation. The occupational pages are where the useful number is.
- Separate your tasks from your title. List last week’s hours and mark each block as producing something to a known specification, or deciding something that needed a person. The ratio is your real exposure.
- Check the sector as well as the role. The same job title in healthcare and in retail faces very different conditions, because healthcare supplies 37 percent of new jobs and retail is shrinking.
- Weigh absolute openings, not only growth rates. Most hiring comes from replacing people who leave, not from growth. A flat occupation with high turnover can offer more openings than a fast-growing small one.
- Notice where the physical constraints are. Solar installers, wind technicians and health aides all appear high on the growth list, and none of them is a language task.
Common mistakes people are making with this data
- Reading exposure categories as automation probability, which BLS explicitly says they are not.
- Concluding that slow overall growth means AI is destroying jobs, when the drivers are largely demographic.
- Choosing a career on percentage growth without checking wages or absolute openings.
- Assuming technical occupations are uniformly at risk, when computer and information research scientists and information security analysts both appear in the top ten fastest growing.
- Treating a ten-year projection as a forecast of next year’s hiring market, which it is not.
What companies are actually hiring for
It is worth separating what employers say from what they advertise. Job postings across most industries have not created large numbers of roles with artificial intelligence in the title. What has changed is the requirements section of ordinary postings.
Marketing roles now ask for experience using these tools in campaign work. Finance roles ask about automating reconciliation and reporting. Operations roles ask about workflow automation. Engineering roles assume it. In each case the company is hiring for the same job it always hired for, with an added expectation that the worker can use the technology competently.
That has two consequences for anyone planning. First, searching for AI jobs is the wrong search. The demand is inside existing roles across every industry, not in a separate category. Second, the evidence employers want is specific: not that you have used a tool, but that you changed a process and can say what it did to time, cost or quality.
The industries hiring hardest for that combination are the ones the projections identify as growing. Healthcare systems need people who can deploy technology into clinical operations without breaking safety. Energy companies building out generation capacity need it in planning and maintenance. Professional services firms are selling exactly this capability to their own clients, which is why that sector is growing 8.6 percent.
Where the Genuine Opportunity Sits in 2026
Pulling the threads together, four clusters stand out.
AI infrastructure and energy. The compute build-out is physical. Electricity generation and distribution, data centre construction, cooling, and the trades that support all of it are growing because of AI rather than despite it.
Healthcare across every level. From nurse practitioners at the top of the wage scale to home health aides at the bottom, this is where the volume is. Ageing populations are the most predictable trend in the entire projection.
Data and security. Data scientists, computer and information research scientists and information security analysts all appear in the fastest-growing list, at wages well above the median.
Roles that combine domain expertise with AI fluency. These do not appear as separate occupations because classification lags reality, but the demand is visible in the professional services growth figure. The person who understands both the business process and what the technology can actually do is the one deciding how it gets deployed.
Building the Fluency That Sits Behind the Growth
Almost every growth area above has the same requirement underneath it. Someone has to decide where these systems belong in a process, design the check that catches their failures, and explain the result to people who will act on it.
That capability is not a job title, which is why it does not appear in any projection table. It is a layer that raises your value inside whatever occupation you are already in, and it is learnable in weeks rather than years. What it takes is understanding which tasks to delegate, how to specify them so the output is checkable, where the predictable failure modes are, and how to build a review step around them. Doing that in a structured sequence produces judgement rather than habits, and pairing it with a certificate makes the capability visible to an employer rather than only useful to you. If you want a structured route in, explore Coursiv AI lessons and check current plan details on the official site.
FAQ
Is AI destroying jobs in 2026?
Which jobs are growing fastest?
What do the AI exposure categories mean?
What should I do with this information?
Your Next Step
Go to the occupational projection for your own job and read the number, then spend twenty minutes marking last week’s calendar into work that followed a specification and work that required a judgement. Most people find the second column larger than they expected, and it is the honest measure of where they stand. Whatever the ratio, the single highest-return move available in this market is becoming the person who decides how these tools are used in your team, because that role exists in every one of the growing sectors and is currently filled almost nowhere.