AI is changing jobs faster than it is eliminating them. The best evidence in 2026 shows no economy-wide job losses from AI, but it does show three clear shifts: fewer entry-level openings in the occupations most exposed to AI, shrinking demand for routine office work, and higher pay for people who can work with AI tools. The future of jobs depends less on whether a role is "exposed" and more on which tasks inside it AI takes over.
The numbers behind that summary come from payroll records, government projections and employer surveys, and they do not all agree. Stanford researchers find that employment for 22 to 25 year olds in highly AI-exposed occupations is about 19 percent below where it would otherwise be. Yale's Budget Lab finds that its results "do not yet show clear indication of labor market effects of AI" across the US workforce as a whole. Both can be true, and this guide explains why, which jobs are shrinking and growing, and what workers and small business owners can do about it.
Will AI replace jobs, or change them?
Most research measures AI exposure: the share of a job's tasks that AI could assist with or perform. Exposure is not the same as replacement. An exposed job can become more productive and better paid, or it can lose the tasks that justified hiring someone.
The International Monetary Fund estimates that almost 40 percent of jobs worldwide are exposed to AI, rising to about 60 percent in advanced economies, 40 percent in emerging markets and 26 percent in low-income countries. It expects roughly half of the exposed jobs in advanced economies to benefit from AI and the other half to see lower demand, lower wages or reduced hiring. A job is a bundle of tasks, and AI rarely takes the whole bundle. What matters is whether the tasks left over still add up to a job someone will pay for.
What the data shows so far in 2026
There is no economy-wide disruption yet
Yale's Budget Lab tracks how the mix of occupations in the US has shifted since ChatGPT launched in late 2022 and compares AI-exposed occupations with unexposed ones. Its 2026 analysis found no clear AI effect on overall employment or unemployment. The change in the occupational mix has been no faster than it was when personal computers and the internet spread through workplaces.
Entry-level workers in exposed jobs are the exception
The Stanford Digital Economy Lab studies payroll records from ADP, the largest payroll processor in the United States. Its August 2026 update covers November 2022 to June 2026 and reports three things:
- Employment for workers aged 22 to 25 in highly AI-exposed occupations is about 19 percent below the level implied by less exposed peers, up from 15 percent a year earlier.
- Experienced workers in the same occupations show no comparable gap.
- The declines are concentrated where AI is used to automate tasks. Where AI is used to complement workers, employment is flat or rising.
The authors describe these as "descriptive patterns, not causal estimates". Higher interest rates and a slowdown in technology hiring affected the same group over the same period, so AI is a likely contributor, not a proven cause.
Most small businesses have not adopted AI
The US Census Bureau asks businesses every two weeks whether they use AI to produce goods or services. Between December 2025 and May 2026 the answer hovered between 17 and 20 percent. Among firms with 250 or more employees it was 37 percent; among firms with fewer than 20 employees it stayed under 20 percent and did not change significantly. Much of the effect on jobs is still ahead, because most employers have not yet changed how work gets done.
Which jobs are most at risk from AI?
The US Bureau of Labor Statistics now factors AI into its ten-year employment projections. Its 2024 to 2034 projections, published in July 2026, expect AI-driven productivity gains to reduce demand for administrative support roles while increasing demand for computer and mathematical ones.
| Occupation | Projected change, 2024 to 2034 | Jobs |
|---|---|---|
| Procurement clerks | Down 8.7 percent | 5,400 fewer |
| Legal secretaries and administrative assistants | Down 5.8 percent | 9,000 fewer |
| Customer service representatives | Down 5.5 percent | 153,700 fewer |
| Claims adjusters, examiners and investigators | Down 5.1 percent | 18,200 fewer |
| Medical transcriptionists | Down 4.9 percent | 2,200 fewer |
| Secretaries and administrative assistants (general) | Down 1.6 percent | 30,800 fewer |
| Software developers | Up 15.8 percent | 267,700 more |
| Operations research analysts | Up 21.5 percent | 24,100 more |
| Information security analysts | Up 28.5 percent | 52,100 more |
| Data scientists | Up 33.5 percent | 82,500 more |
The declining roles share a profile: the work arrives as text, follows written rules and happens in high volume. Those are the same properties that make a process suitable for automation, as we explain in our guide to agentic AI for business. The declines are also gradual. A 5.5 percent fall over ten years still leaves more than 2.5 million customer service jobs in the United States.
Which jobs is AI creating?
New technology has a long record of creating work that did not exist before. Research led by MIT economist David Autor found that about 60 percent of US employment in 2018 was in job titles that did not exist in 1940. The open question is whether AI creates new work as fast as it removes old tasks.
Three signs of where the growth is so far:
- Technical roles. Data scientists, security analysts and software developers lead the BLS projections above. AI changes how developers work without removing the need for them, as we cover in how AI is changing web development.
- Jobs that require AI skills. PwC's 2026 Global AI Jobs Barometer, based on job advertisements in 27 territories, found that workers with AI skills earn an average wage premium of 62 percent, and that postings requiring those skills grew far faster than the overall job market.
- Work AI cannot do. The World Economic Forum expects the largest growth in absolute numbers to come from frontline and care roles such as farmworkers, delivery drivers, construction workers and nurses, driven by demographics and the energy transition more than by technology.
How many jobs will AI replace by 2030 and 2035?
The headline forecasts differ because they measure different things over different periods.
| Source | Scope | Forecast |
|---|---|---|
| World Economic Forum, Future of Jobs Report 2025 | Global, 2025 to 2030, all trends including AI | 170 million jobs created, 92 million displaced, a net gain of 78 million |
| McKinsey Global Institute, September 2026 | United States, to 2035 | Automation reduces demand by 36 million jobs, growth creates demand for 40 million, and about 11 million workers need to change occupation |
| International Monetary Fund, 2024 | Global, no end date | 40 percent of jobs exposed, about half of those in advanced economies at risk of lower demand |
| US Bureau of Labor Statistics, 2026 | United States, 2024 to 2034 | Single-digit percentage declines in specific office roles, double-digit growth in technical ones |
None of these predicts mass unemployment. All of them predict movement, and the McKinsey figures show who carries the cost of it. More than three quarters of the workers who may need a new occupation are in office and administrative support, retail and sales, or transportation and logistics. Lower-wage workers are 7.6 times more likely than higher-wage workers to need a change, and only about one in seven has a direct route into a new role without retraining or a pay cut.
Why are entry-level jobs under the most pressure?
A large study of customer support agents helps explain the Stanford result. When agents were given an AI assistant, productivity rose 14 percent on average and 34 percent for the newest, least experienced workers, with little gain for the most experienced. The assistant had effectively captured what top performers knew and handed it to beginners.
That is good news for a junior employee who has a job. It also means the tasks juniors used to learn on, such as first drafts, routine tickets and basic research, are the tasks AI now does well. Our reading, which goes beyond what the studies themselves claim, is that employers are responding by hiring fewer beginners, not by dismissing experienced staff. The risk for any business is the pipeline: the senior people of 2035 are the juniors who are not being hired in 2026.
What skills will matter most?
Employers surveyed by the World Economic Forum expect 39 percent of workers' core skills to change by 2030. Analytical thinking remains the skill they value most, while AI and big data, cybersecurity and technological literacy are the fastest growing. For an individual worker, that translates into five practical steps:
- List your tasks, not your job title. Mark each as rule-following, judgment or relationship work. The first group is where AI arrives first.
- Learn the AI tools used in your field. Our guide to using AI at work covers the everyday tasks where they save time and where they make mistakes.
- Become the person who checks the output. Reviewing AI work needs domain knowledge, and it is the part of the process a business cannot skip.
- Move toward the customer or the decision. Work that involves trust, negotiation or accountability is the slowest to automate.
- Keep evidence of results. A record of what you delivered matters more as routine tasks stop being a way to show your ability.
What should small business owners do?
For a small employer the question is less about replacing people and more about what a team of the same size can now handle.
- Audit tasks before roles. Time the repetitive work for a week: inquiry replies, data entry, scheduling, invoice handling.
- Automate what is frequent and rule-based. Start with one process, with a person approving the output until it has earned trust. This is the work our AI and business automation service is built around.
- Train the staff you have. The same WEF survey found that 85 percent of employers plan to upskill their workforce, which is cheaper than hiring for AI skills at a 62 percent premium.
- Keep hiring juniors, and change what they do. Give them review, customer contact and exceptions earlier, with AI handling the routine work they would once have started on.
- Be clear with your team. Say which tasks you plan to automate and what people will do with the time. Uncertainty costs you your best staff first.
If you are not sure which processes are worth automating, an IT consulting review is a lower-risk first step than buying a tool.
Frequently asked questions
For most people it will change the job before it removes it. Current evidence shows no broad job losses from AI, but roles made up mostly of routine, text-based, rule-following tasks are projected to shrink over the next decade, and entry-level hiring in those roles has already slowed.
Jobs that depend on physical work in varied settings, in-person care, or accountability for decisions. Skilled trades, nursing, construction and roles built on client relationships are among those forecast to grow. No job is fully protected, because AI can still change the administrative part of almost any role.
The World Economic Forum expects 92 million jobs to be displaced worldwide between 2025 and 2030 and 170 million to be created, a net gain of 78 million. Those figures cover all major trends, including AI, demographics and the energy transition, so AI accounts for only part of both numbers.
Not at a scale that shows up in national employment data. Yale's Budget Lab finds no clear AI effect on US employment so far. The measurable effect is slower hiring of young workers in AI-exposed occupations, which Stanford researchers put at about 19 percent as of mid 2026.
Rarely as a first move. Most small firms gain more by automating repetitive tasks and using the time for sales, service and work that was not getting done. Cutting too far also carries risk: AI output still needs someone who understands the business to check it.
A one-hour exercise for your own team
Pick one role in your business and write down everything that person did last week. Next to each item, note how often it happens, whether it follows written rules, and what a mistake would cost. The items that are frequent, rule-based and cheap to get wrong are your automation candidates. Everything else is the job that remains, and it is usually the part customers pay for. If you would like a second opinion on that list, our automation team can go through it with you on a free call.
