By Salary by Career 
An insurance underwriter in 2035 may still have an insurance underwriter job. What fills the workday, though, could look noticeably different.
US underwriter employment is projected to stay essentially flat at about 87,000 full-time equivalents through 2035. Less time could go into gathering information and processing routine data.
More time could go into:
- Assessing risk
- Checking AI recommendations
- Making judgment calls
- Explaining decisions to clients or colleagues
Nobody disappeared in that example. Neither did change.
A person could keep the same title, sit at roughly the same desk, earn a paycheck through the same occupation, and still discover that a large portion of what once defined the job has moved elsewhere.
Such a distinction matters because much of the debate around AI employment starts with a dramatic question: How many jobs will disappear?
A quieter question may become relevant much sooner. How much of your current job will still exist inside your future job?
AI Replaces Tasks Before It Replaces Jobs
Jobs sound neat when written on a business card. Actual work is messier.
An accountant does not perform one task called accounting for eight hours. A teacher does not spend an entire day doing one activity called teaching. Most occupations contain a collection of responsibilities, some repetitive, some analytical, some social, and some difficult to describe until somebody stops doing them.
Research by Meltwater shows that 75% of LinkedIn content cited in AI-generated answers comes from individual users rather than company pages, suggesting that personal posts play a substantial role in shaping the information AI systems present to their users.
Researchers compared AI systems used by companies with tasks associated with individual occupations. Once an AI system could perform something closely resembling a certain task, that task became significantly less likely to appear in later job postings at the company.
That does not automatically mean the associated occupation vanished.
Imagine an employee with ten meaningful responsibilities. AI takes over two. Management now has several options. Staff numbers could fall, or employees could devote more time to the other eight responsibilities. Companies could also add responsibilities that previously received little attention because nobody had enough hours in the day.
After accounting for task reallocation and company growth, AI’s net effect on the mix of occupations was relatively modest during the study period. AI still explained about 14% of variation in occupational employment-share growth.
14% is substantial enough to notice. It is not substantial enough to reduce a complicated labor market to a story about machines simply replacing people.
One important limitation sits behind those results. Most observations came before generative AI became widely available. Earlier AI systems often handled narrower activities than today’s tools, so those findings cannot tell us exactly how newer technology will affect employment.
Still, one lesson travels reasonably well across technologies.
A job is a bundle of tasks. Automating part of that bundle is not automatically the same thing as eliminating its owner.
So Far, AI Is Changing Work More Clearly Than Employment
Most companies using AI reported no overall change in employment tied to adoption. Among businesses that did report a change, increases in employment were more common than decreases.
Changes inside jobs were easier to spot.
Among companies already using AI:
- 44% said AI supplemented or enhanced work employees were already performing.
- 10% said AI performed a task previously completed by an employee.
- 11% said AI introduced a task that nobody at the company had previously performed.
- 44% sounds comforting at first. Then comes a small complication.
Keeping your job and keeping the same job are not identical outcomes.
Someone who once produced weekly reports may now spend more time checking AI-generated reports. A customer support worker may handle fewer routine questions and more frustrated customers whose problems escaped automated systems. An analyst may produce fewer initial calculations and spend more time deciding which outputs deserve trust.
Employment can look stable while daily work changes considerably.
Companies adopting AI have also tended, on average, to become more productive and expand employment even while demand weakened for certain highly exposed occupations.
No rule says that pattern has to continue indefinitely. Generative systems may improve faster than earlier tools, employers may discover new ways to reorganize work, and labor demand could respond differently as adoption spreads.
Current evidence simply makes today’s situation harder to summarize with the familiar formula of AI arrives, worker leaves. So far, AI adoption looks at least as much like job redesign as job destruction.
Career Reinvention Could Affect Far More People Than Layoffs
According to McKinsey report, more than 70% of US workers could experience some degree of occupational reinvention by 2035, with new or reassigned tasks consuming more than 15% of their working time.
About 25% could see more than 30% of task time reallocated. Another 47% could experience changes covering 15% to 30% of their working time.
None of those numbers requires a worker to lose a job.
Consider cashiers.
Furthermore, McKinsey states that employment could decline by between 2.9 million and 2.4 million workers by 2035. Remaining cashiers could spend less time processing transactions and more time helping customers and overseeing automated systems.
Consider insurance underwriters again.
Headcount could stay around 87,000, yet less time may go into collecting data while more goes into risk judgment and AI oversight.
Truck driving provides an even more interesting example because employment could grow.
Driver employment could increase between 1.3 million and 1.5 million even as paperwork and data-entry duties shrink. Driving, cargo handling, and other physical responsibilities could account for more of each workday. Growth, stability, and decline can therefore produce the same underlying experience: substantial changes in what workers actually do.
Counting jobs alone misses that.
An occupation could add 200,000 workers and still demand a noticeably different skill set. Another could lose 500,000 positions yet leave millions of people employed in redesigned roles.
For workers planning a career, that distinction may matter more than a headline about total employment.
Valuable Skills May Shift Toward Judging Work
AI fluency has moved quickly through job requirements.
Between 2022 and 2026, occupations demanding AI fluency increased roughly elevenfold. Demand for adaptability increased fivefold. Demand for resilience, curiosity, and willingness to learn roughly tripled.
Those figures can tempt employers and workers toward an overly simple conclusion: learn how to use AI tools, and everything will work out.
Reality is more demanding. Using AI may become fairly ordinary. Judging its output is harder.
AI fluency is about to become as common as knowing how to use email. It’s not a specialist skill, but a baseline skill everyone in an organization will need. Kerri M. Roberts explores what that means in practice and how businesses can build it without a large investment in Don’t Waste the Chaos.
Junior radiologists did not simply hand the work to software and leave early. Human effort moved toward reviewing drafts, spotting mistakes, correcting them, and deciding what deserved clinical trust.
Teachers have reported a similar shift. AI use saved an average of 5.9 hours per week, with much of that time coming out of administrative duties and grading.
Saving nearly six hours is meaningful. What happens next is the more interesting part.
Teachers could spend that time preparing lessons, working directly with students, reviewing difficult cases, or doing another task their employer now expects because six hours suddenly became available.
Productivity gains have a habit of finding new owners.
Asking a machine the right question can help. Knowing when its answer is nonsense can help even more.
Watch the Job Description, Not Only the Unemployment Rate
Nobody yet has a clean answer to the biggest AI employment question.
Current evidence cannot tell us exactly how many jobs advanced AI systems will eventually eliminate. Systems are improving, companies are still experimenting, and adoption has not reached every occupation equally.
Clearer evidence exists on something happening sooner.
Careers can change substantially without disappearing.
Tasks can move to software. Workers can inherit new responsibilities. Skills once considered secondary can become valuable. Jobs can grow in number while changing internally.
Stable employment can hide difficult adjustments for individuals whose old skills suddenly buy less in the labor market.
Such changes may be easy to miss if attention stays fixed on unemployment.
A stable unemployment rate can tell us that people still have jobs. It cannot tell us how much those jobs have changed. Job descriptions may give away the answer first.