What Jobs Will AI Replace? What the Research Says in 2026, Role by Role

Lists of jobs AI will replace are easy to find. The research behind them is more careful. The International Labour Organization says transformation, not replacement, is the most likely outcome for exposed jobs. A March 2026 study that matches AI usage data to US labor statistics finds no systematic rise in unemployment among the most exposed workers.

That does not mean nothing is happening. Some tasks are already moving to software, some roles are shrinking, and entry-level hiring into exposed jobs may be slowing. Here is what the evidence says, and how to read it for your own job.

Does AI replace jobs or tasks?

AI replaces tasks first. The main exposure studies measure AI's reach task by task, then add the tasks up into occupations. A job with many AI-ready tasks is "exposed," which is not the same as "replaced." What happens next depends on demand, regulation and how employers reorganize the work.

The authors of Microsoft Research's "Working with AI" study put it plainly: AI's effects on employment and wages "will depend on hard-to-predict business decisions." Their example is the ATM, which automated a core task of bank tellers "but led to an increase in the number of bank teller jobs as banks opened more branches at lower costs."

That story has a second chapter. The World Economic Forum's Future of Jobs Report 2025 now lists bank tellers among the fastest-declining roles. Task automation can support a job for years and still reshape it in the end.

Which jobs are most exposed to AI?

The most exposed jobs are information-heavy: translators, writers, customer service representatives, programmers, data entry and clerical roles. Five major sources measure exposure in different ways, from real AI usage to employer surveys to government projections, and they agree more than they differ. Customer service and clerical work appear near the top of almost every list.

Two findings stand out. First, exposure now reaches well-paid work: in Anthropic's data, workers in the most exposed quarter earn 47% more on average than workers with no exposure, and are almost four times as likely to hold a graduate degree.

Second, the early warning sign is in hiring, not layoffs. In the same Anthropic study, the rate at which 22 to 25 year olds started jobs in highly exposed occupations averaged 14% lower after ChatGPT's release than in 2022, a result the authors call "just barely statistically significant." There was no such drop for workers older than 25.

Which jobs are hardest for AI to replace?

The jobs AI can't replace today are physical, hands-on and in-person: care work, construction and maintenance, food service and safety roles. Microsoft's lowest-scoring groups involve "physically working with people, operating machinery, and other manual labor." In Anthropic's data, 30% of workers do jobs where AI covers none of the tasks.

• Hands-on care: home health aides and nursing assistants score 0.04 in Microsoft's study, among the lowest of any group.
• Building and outdoor work: construction trades workers (0.07), grounds maintenance workers (0.04) and forest and conservation workers (0.03).
• Food, service and safety: Anthropic's zero-coverage group includes cooks, motorcycle mechanics, lifeguards, bartenders, dishwashers and dressing room attendants.
• Growing frontline roles: the WEF expects the largest absolute job growth by 2030 for farmworkers, delivery drivers and construction workers, plus nursing professionals and secondary school teachers.

Inside knowledge jobs, the hardest parts to replace are judgment and relationships. When Anthropic surveyed Claude users about tasks AI would never do, the most common answers stressed the "judgment, contextual awareness, and situational reasoning" their work requires, and the relational side of the job, "building trust and managing people." That sample skews toward heavy AI users, so treat it as a signal, not a census.

Will AI replace your role? A role-by-role look

For the eight professions below, AI is changing tasks in every one, but BLS still projects stable or growing employment for all of them. The clearest pressure is on paralegals, projected flat, and personal financial advisors, whose projected growth fell from 10% to 1% between the last two BLS editions. Each linked page has the detail and three real practice questions.

Read the verdicts together and a pattern appears. Where BLS expects growth despite automation, as for accountants and lawyers, it expects people to spend more time on advice, analysis and clients. Where routine work is most of the job, as for paralegals, it expects demand to flatten.

How can you judge your own role?

Judge your tasks, not your job title. Two people with the same title can face very different exposure, depending on how their week is split between routine information work and work that needs presence, accountability or trust. Four questions give you a quick read.

1. How much of your week is drafting, summarizing, searching or answering standard questions? That is the information work where AI applies most.
2. Which tasks need you physically present? Hands-on work is the least exposed in both usage studies above.
3. Which outputs do you sign, own or answer for? Responsibility does not transfer to a tool, so the review and sign-off stay with you.
4. Which tasks depend on a relationship? Experienced workers in Anthropic's survey named building trust as something AI cannot replicate.

If most of your week falls under the first question, your job will change fastest. That is a reason to learn the tools now, not a verdict on your career.

What should you do about it?

Build the skills that move with you from tool to tool and job to job: deciding what to delegate, giving AI context, checking its output, protecting data and owning the result. The WEF expects nearly 40% of the skills required on the job to change, and 59 in every 100 workers to need reskilling or upskilling by 2030.

Skills like these come from practice on real work, a little at a time, which is the idea behind AI fluency. For the full list and how teams build them, read AI skills every team needs. Then open the role page closest to your job and try its three practice questions: they come from the same daily sessions kju learners use.

What should team leads do?

Treat AI as a team change, not a tool rollout. In the WEF survey, 77% of employers plan to upskill their workers, 41% plan to reduce their workforce as AI automates certain tasks, and almost half expect to move staff from AI-exposed roles into other parts of the business. Redeployment only works when people have the skills for the work they move into.

Start by mapping which tasks in each role are changing, then train the whole team on those tasks, not a few enthusiasts. kju gives each employee daily AI practice tailored to their role, industry and skill level, and lets you measure progress across every team. See kju for teams, or read how to close the AI skills gap in your organization.