Which Jobs Will AI Replace, and Which It Will Not

No job is completely AI-proof, but few jobs are likely to be fully replaced by AI soon. AI changes work task by task. Jobs made up mostly of routine digital tasks, such as data entry and many clerical roles, face the most pressure. Jobs that combine hands-on work in unpredictable settings, close human interaction, judgment and accountability are more resistant. For most people, AI is likely to change how a job is done before it changes whether the job exists.

Why tasks matter more than job titles

A job is a bundle of tasks. A paralegal researches cases, drafts documents, organizes files, talks with clients and coordinates with courts. AI might speed up some of those tasks, take over others and barely touch the rest.

That is why asking "Will AI replace this job?" often produces misleading answers. The better question is "Which tasks in this job can AI do, and what happens to the job when it does them?"

The U.S. Bureau of Labor Statistics makes this point in its approach to employment projections: new technologies may change the tasks workers perform, "sometimes dramatically," and "may still have no employment impacts."

How researchers frame the question changes the answer:

• A widely cited 2013 Oxford study treated whole occupations as automatable and estimated that about 47% of total U.S. employment was at risk of computerization.
• A 2016 OECD analysis looked instead at the tasks within each job. It estimated that, on average across 21 countries, 9% of jobs were automatable, because many "high-risk" occupations still include substantial work that is hard to automate.

The two studies used different methods and countries, and both predate generative AI, but the gap shows how much framing matters. The International Labour Organization reached a similar conclusion about generative AI in 2025: because most occupations consist of tasks that require human input, "transformation of jobs is the most likely impact."

Exposure, augmentation and automation are not the same

Headlines often blur three very different ideas.

Exposure is a measure of potential, not a prediction. The Bureau of Labor Statistics, which introduced its own AI exposure categories in August 2026, states plainly: "Exposure does not imply job loss, productivity gains, automation probability, or wage effects."

AI-driven job losses generally depend on how many of a job's tasks are automated, and whether demand for the work grows enough to absorb the change. Sometimes cheaper, faster work increases demand. Sometimes it does not.

Real-world use includes both augmentation and automation. When Anthropic analyzed how people used its Claude AI assistant in data from November 2025, it classified 52% of sampled conversations as augmentation and 45% as automation. That reflects one product's users, not the whole workforce, and the split has shifted between Anthropic's reports as its products changed.

What the evidence shows so far

Nearly four years after ChatGPT's release, research points to a mixed picture: little sign of broad job losses, alongside early signs of pressure in specific places.

No broad disruption, yet. The Budget Lab at Yale found no discernible disruption to the broader U.S. labor market in the 33 months after ChatGPT's release, and its tracker, updated in August 2026, still found that measures of AI usage "show no connection to changes in employment or unemployment." Anthropic's economists reported in March 2026 that they found no systematic increase in unemployment for highly exposed workers since late 2022.

Early-career workers may be feeling it first. Stanford Digital Economy Lab researchers, analyzing payroll data, found no evidence of widespread job displacement but reported that employment of workers aged 22 to 25 in AI-exposed occupations was 19% below where it would have been had it kept pace with less-exposed peers, as of June 2026 data. The gap came mainly from reduced hiring rather than more workers leaving jobs, and was concentrated in jobs where AI mostly substitutes for workers. Where AI mostly complements workers, employment was flat or rising. The study is a working paper that has been revised as new data arrived.

In one regional survey, employers were retraining more than cutting. In a New York Federal Reserve survey of businesses in the New York and northern New Jersey region, published in September 2025, only 1% of service firms using AI reported laying off workers because of AI in the previous six months, while just over a third reported retraining workers in response to AI.

AI can raise productivity, unevenly. A study published in The Quarterly Journal of Economics in 2025 found that customer support agents with access to an AI assistant resolved 15% more issues per hour on average, with the biggest gains for less experienced workers. A separate experiment with consultants at Boston Consulting Group found AI users completed 12.2% more tasks, 25.1% faster, on tasks within AI's capabilities. On a task outside those capabilities, they were 19% less likely to reach a correct answer than consultants without AI.

Work likely to change the most

Work made up mostly of routine, rules-based digital tasks faces the biggest changes. Clerical and administrative roles appear at or near the top in analyses by the International Labour Organization, the World Economic Forum and the U.S. Bureau of Labor Statistics.

• The International Labour Organization found in 2025 that clerical jobs have the highest exposure to generative AI. It also found that only 3.3% of global employment falls into its highest-exposure category, and that women are more likely than men to hold those highly exposed jobs.
• The World Economic Forum's 2025 employer survey expects clerical and secretarial roles, including cashiers, ticket clerks, administrative assistants and executive secretaries, to see the largest declines in absolute numbers by 2030. Postal service clerks, bank tellers and data entry clerks are among the fastest-declining roles in percentage terms. These employer expectations reflect all forces reshaping jobs, not AI alone.
• The Bureau of Labor Statistics' 2025 to 2035 projections, released in August 2026, expect office and administrative support occupations to decline 4.0%, shedding about 752,100 jobs, partly because of automation tools including those powered by AI. In its previous 2024 to 2034 projections, it also expected employment of medical transcriptionists to decline because AI can recognize speech and transcribe audio.

The tasks under the most pressure include:

• entering, checking and moving data between systems
• transcribing and summarizing recordings and documents
• answering routine, repetitive customer questions
• scheduling and basic coordination
• producing first drafts of standard documents and reports

"Declining" does not mean disappearing. A projected decline over a decade means fewer openings and changing duties, not that everyone in those roles will be out of work.

Work more resistant to automation

Some kinds of work are much harder for AI to take over. The more a job depends on the factors below, the more resilient it tends to be.

The evidence behind these factors:

• Physical, unpredictable work. Even the 2013 Oxford study that estimated high automation risk identified unstructured work environments as a key barrier to automation. Separately, the World Economic Forum's employer survey expects frontline roles such as farmworkers, delivery drivers and construction workers to add the most jobs by 2030, and projects significant increases for care roles such as nursing professionals. Those projections reflect demand trends, not only resistance to automation.
• Social skills. Research by economist David Deming found that U.S. jobs requiring high levels of social interaction grew by nearly 12 percentage points as a share of the labor force between 1980 and 2012.
• Review, regulation and client preference. Bureau of Labor Statistics case studies point to the continuing need for human reviewers to catch AI errors, regulations requiring engineers to review and approve work, and client demand for human counsel in complex financial matters.
• Demand for AI-related work. The Bureau of Labor Statistics projects employment of data scientists to grow 35% from 2025 to 2035, and the World Economic Forum expects AI and machine learning specialists to be among the fastest-growing roles in percentage terms.

These patterns are not guarantees. Generative AI can now perform some creative and language tasks that earlier research considered safe, and robotics continues to improve. Resilience is about which tasks a job depends on, not the job title alone.

How many jobs will AI replace?

There is no reliable single number. Published figures often measure different things, and many describe exposure rather than job losses.

The last estimate illustrates how assumptions matter. In the peer-reviewed version of that research, published in Science in 2024, the authors estimated that about 1.8% of jobs could have over half their tasks affected by language models with simple interfaces, rising to just over 46% once current and likely future software built on top of those models is taken into account.

The Bureau of Labor Statistics sums up the state of knowledge: "Developments in AI are proceeding rapidly, and the uncertainty about potential impacts remains very high."

Will AI take my job? A practical way to think about it

Concern is growing among young adults: a Pew Research Center survey from June 2026 found that 73% of U.S. adults under 30 believe AI will lead to fewer jobs. For most people, AI is more likely to change parts of their job than to eliminate it, but the risk varies by role, industry and career stage. Instead of relying on lists of "safe" and "doomed" jobs, look at your own work.

1. List your main tasks. Write down what you actually spend your time on in a typical week.
2. Sort them. Mark tasks that are routine and digital, such as drafting, data handling and scheduling. Then mark tasks that depend on physical presence, relationships, judgment or responsibility.
3. Ask how AI would be used. Would AI make you faster at a task while you stay in charge, or could it complete the task without anyone checking?
4. Consider demand. If AI makes your work cheaper or faster, is there more of it to be done, or will fewer people be needed?
5. Watch hiring, not just layoffs. Early evidence suggests AI's effects may show up first in reduced hiring of young workers.

Then act on what you find:

• Learn to use AI in your own field. Workers who know how to use AI well are better placed as tasks change. How to Learn AI Skills offers a starting path.
• Strengthen the parts of your work AI handles poorly: judgment, relationships, hands-on expertise and responsibility for outcomes.
• Move toward tasks that direct and check AI output. As more work is drafted by AI, the ability to review it well becomes more valuable.
• Start using AI in low-risk ways now. How Do You Use AI? covers practical first steps.