What Jobs Will AI Replace? Which Jobs Are Safe in 2026

Two statements about AI and jobs are both true at the same time, and the tension between them explains almost all the confusion.

Statement one: Goldman Sachs says two-thirds of US occupations are exposed to AI automation and 300 million jobs globally could be affected.

Statement two: The WEF Future of Jobs Report 2025 projects AI will create 78 million more jobs than it destroys by 2030.

Both are correct. The panic and the optimism just talk past each other because they’re answering different questions. The question worth answering is the specific one: which jobs, which tasks, and why?

What the Research Actually Shows

The Goldman Sachs economists who published that “two-thirds exposed” figure were careful about what they meant. For most roles, AI could automate 25–50% of tasks within the role — not replace the role entirely. Direct near-term displacement risk applies to about 2.5% of US employment. A broader 6–7% displacement is projected over a full decade of AI adoption.

The WEF surveyed 1,000+ employers covering 14 million workers across 55 economies for their 2025 report. Their headline: 92 million jobs displaced by 2030, 170 million new jobs created, net gain of 78 million. But the distribution matters: 40% of employers anticipate reducing headcount in roles where AI can automate tasks, while 86% expect AI to transform their operations significantly by 2030.

Korn Ferry’s 2025 Talent Acquisition Trends Report adds the employer intention layer: 43% of companies plan to replace roles with AI. Their specific targets are operations and back-office functions (58% of those companies) and entry-level positions (37%).

Indeed’s AI at Work Report 2025, analyzing 53.5 million US job postings, found that 26% are “highly transformed” by AI, 54% moderately transformed, and 20% barely affected.

The Jobs Under Most Pressure

The WEF’s data on fastest-declining roles through 2030 is specific enough to be useful:

Administrative and clerical roles:

• Data entry clerks: projected 26–34% decline
• Bank tellers and related clerks: 31–35% decline
• Administrative assistants and executive secretaries: 6.1 million net job losses
• Accounting, bookkeeping, and payroll clerks

Customer-facing routine roles:

• Cashiers and ticket clerks: 13.7 million net losses globally
• Customer service and call center workers

Other vulnerable categories:

• Claims adjusters and insurance examiners
• Paralegal support — Goldman Sachs found 44% of legal tasks are automatable
• Some graphic design work (0.36 million net loss projected)

The pattern across all of them: high volume, repetitive cognitive work, rule-based decision-making, document processing. Large language models are particularly good at this kind of work. They’re not particularly good at dealing with a flooded basement, a distressed patient, or a client relationship that’s about to collapse.

McKinsey’s Generative AI and the Future of Work research projects that office support occupations will see about 18% demand reduction and customer service about 13% by 2030 — while healthcare is projected to add 5.5 million jobs and STEM demand will grow 23%.

The Jobs That Are Safer

The same research consistently identifies another category: roles that resist automation not because they’re complex in a technical sense, but because they require things AI genuinely can’t do.

Healthcare and care work: Indeed found 68% of nursing skills fall into “minimal transformation” categories. The Bureau of Labor Statistics projects nurse practitioners will grow 45.7% through 2032, physician assistants 27.6%, and mental health counselors 22.1%. These projections account for AI. The reason isn’t that healthcare is technologically backward. It’s that physical touch, emotional presence, and clinical judgment in genuinely unpredictable situations can’t be reliably automated.

Skilled trades: Electricians, plumbers, HVAC technicians, and construction workers operate in constantly changing physical environments. AI has no hands. It can’t diagnose a wiring problem in a 1960s house, or figure out why the water pressure is wrong in a specific apartment. The BLS projects 663,000+ construction and extraction job openings per year through 2033.

Education and social work: Teachers, social workers, and child care workers build relationships that require sustained human presence. These relationships are the core of the work, not a peripheral feature.

Complex management: The WEF data shows a sharp seniority effect: AI can handle 53% of a junior market research analyst’s tasks versus 9% for their manager. The more a role involves navigating ambiguity, managing competing priorities, and making calls that require judgment and context, the less AI can substitute for it.

The WEF’s fastest-growing jobs list includes software developers, farmworkers, nursing professionals, building trades workers, operations managers, and social workers — a mix that reflects exactly this pattern.

The Most Important Distinction: Exposed vs. Eliminated

This is where most AI-jobs coverage goes wrong.

Indeed classified fewer than 1% of skills across 53 million job postings as facing “full transformation” — where AI can handle the task entirely. The overwhelming majority of exposure falls into “hybrid transformation,” where AI handles portions of the work while humans handle the rest.

Software developers are the clearest example. They rank among the most AI-exposed occupations: 81% of their listed skills fall into hybrid transformation categories. They also appear on the WEF’s fastest-growing jobs list for 2030. High exposure to AI transformation doesn’t mean a profession is disappearing. It means it’s changing, and workers who change with it will be in demand.

The real risk isn’t being in an “exposed” job category. It’s being locked into the most routine, most automatable slice of that category with no path toward the judgment-heavy parts AI can’t touch.

The Entry-Level Problem

One pattern across multiple studies deserves specific attention: AI is compressing career entry points.

Goldman Sachs reported in April 2026 that AI is cutting approximately 16,000 US jobs per month on a net basis, with workers under 30 facing disproportionate impact. New graduate unemployment reached nearly 10% by late 2025. An ADP/Stanford study found software developers aged 22–25 saw nearly 20% fewer positions since 2022 — not because the field is declining, but because AI handles the work that entry-level developers used to do to build experience.

Korn Ferry explicitly warned about this: eliminating entry-level roles to cut costs “opens the door to a long-term leadership crisis.” The entry-level position is where people learn to become mid-level and senior. Remove it and you’ve cut the pipeline.

For anyone starting a career or trying to break into a new field, this raises the stakes for every application. Fewer openings, more competition, higher bar. If you’re looking for an edge, our piece on which skills employers want in an AI-driven market covers the specifics. And our 2026 job market crisis guide addresses the tactics for standing out when the competition is steep.

What to Do With This

The practical upshot from the research:

Understand your exposure. Look at the tasks you actually do most. High-volume, rule-based, easily documented tasks are the ones AI targets first. If that describes most of your current role, that’s useful information.

Move toward judgment-heavy work. The parts of any job that involve weighing competing priorities, managing relationships, and making calls with incomplete information are the parts AI can’t reliably handle. Get more of your time into those areas.

Make your resume show outcomes, not tasks. “Processed 200 invoices weekly” is automatable. “Reduced invoice processing errors by 34%, saving $180K annually” demonstrates judgment. ResuFit helps you frame your experience in those terms, and our resume analyzer checks whether your current resume would make it past AI screening.

For the deeper picture on what AI is actually doing to career paths — including the Superworker concept and why it’s incomplete — our piece on the Superworker Myth is worth reading alongside this one.