Which Jobs Is AI Changing the Most? A Task-by-Task Look for 2026

“Will AI replace my job?” is one of the most common questions people type into a search bar right now, and the honest answer is more useful than a yes or no. AI rarely takes over an entire job. It takes over pieces of jobs: the first draft, the data lookup, the routine reply, the summary of a long document.

That distinction matters. If you think in whole jobs, the future looks like a cliff. If you think in tasks, it looks like a remodel, and you get a say in how your role gets rebuilt. This guide looks at what the official data actually shows, which kinds of work are changing fastest, and a practical way to map your own exposure.

The short answer: AI changes tasks before it changes jobs

Every job is a bundle of tasks. A paralegal researches case law, drafts documents, organizes files, schedules depositions and talks to clients. Generative AI is already strong at some of those (summarizing and drafting) and weak at others (judgment calls, relationships, physical work, accountability).

Usage data backs this up. Anthropic’s first Economic Index report, published in February 2025 and based on conversations with its Claude assistant, mapped usage to the U.S. government’s O*NET task database. It found that about 36% of occupations had AI used for at least a quarter of their tasks, but only about 4% used it for at least three-quarters of their tasks. It also found more augmentation (people working with AI, iterating and checking) than pure automation, at roughly 57% versus 43%. The company notes the data covers only its own product, so it is not a full picture of AI use.

The pattern is clear even with that caveat: AI is spread thinly across many jobs, and deep in only a few.

Key takeaway: Stop asking whether your job title is safe. List the tasks you do each week and ask which ones a tool could draft, check or complete. That list tells you far more than any “jobs at risk” ranking.

What the official US data says

The U.S. Bureau of Labor Statistics (BLS) has started building AI into its 10-year employment projections, and it is deliberately cautious about it.

In a February 2025 Monthly Labor Review article, BLS economists walked through case studies from the 2023–33 projections. They projected declines for several occupations where AI and automation can take over core tasks, while expecting growth to continue in others:

The average growth rate for all occupations in that cycle was 4.0%. BLS says it applies AI adjustments “conservatively, where there is convincing evidence for a change,” and that uncertainty about AI’s impact remains very high.

The newer 2025–35 projections, released August 27, 2026, point the same way at a broader level. BLS projects total U.S. employment to grow 3.5% over the decade. Office and administrative support is the fastest-declining major group, projected to fall 4.0% and lose about 752,100 jobs, and BLS cites generative AI and automation as factors that may reduce demand in some office, sales, arts and media roles. At the same time, it projects fast growth for data scientists (+34.6%) and nurse practitioners (+41.0%).

BLS now publishes AI exposure tiers

Alongside the 2025–35 projections, BLS introduced a new product that sorts 831 detailed occupations into 4 relative AI exposure tiers: Low, Moderate, High and Very high. It blends “theoretical” exposure (how well AI capabilities match a job’s tasks) with “observed” exposure from real AI usage data. BLS is explicit that high exposure does not mean job loss, and low exposure does not mean a job is safe from change. The tiers also do not separate automation from augmentation. Treat them as a starting point for your own task review, not a verdict.

The kinds of work AI is changing the most

Across the BLS case studies, usage research and employer surveys, the same types of tasks come up again and again.

Routine customer communication

Answering common questions, resetting passwords, tracking orders and routing tickets are now handled by AI assistants in many companies. Human agents increasingly handle the escalations: angry customers, unusual cases and anything involving money or judgment. The job does not vanish, but entry-level volume shrinks and the remaining work gets harder.

Data entry, processing and clerical work

The World Economic Forum’s Future of Jobs Report 2025, based on a survey of more than 1,000 large employers, named postal service clerks, bank tellers and data entry clerks among the roles employers expect to decline fastest by 2030. These jobs are heavy on structured, repetitive information handling, which software has been absorbing for years and AI is accelerating.

Writing, summarizing and research

Drafting emails, reports, marketing copy and first-pass legal or policy documents is one of the most common uses of generative AI. Writers, editors, paralegals and analysts are not out of work, but the value shifts toward editing, fact-checking, voice and deciding what is worth writing in the first place.

Software development

Coding is one of the most AI-exposed activities, yet BLS still projects strong growth for software developers. The reason: AI makes code cheaper to produce, which increases demand for software, while someone still has to design systems, review output and own the results. We cover this shift in detail in how AI coding assistants are changing software teams.

Image-based assessment

Auto damage appraisal and some inspection work are changing because AI can estimate from photos. This is a good reminder that exposure is not only about text.

Jobs AI is changing least, for now

Work that combines physical skill, unpredictable environments, hands-on care or high-stakes trust is changing more slowly. BLS’s fastest-growing list for 2025–35 is full of these: nurse practitioners, solar photovoltaic installers and wind turbine service technicians. Skilled trades, healthcare delivery, early childhood education and on-site technical work all involve tasks that current AI cannot do on its own.

That said, AI still touches these jobs at the edges: scheduling, documentation, diagnostics support and customer communication. A plumber who uses AI for quotes and invoices still does the plumbing.

How to map your own AI exposure in 5 steps

You do not need a research team to figure out how AI affects your role. Block out an hour and do this:

1. List your recurring tasks. Write down 15 to 25 things you actually do in a typical month, not your job description.
2. Estimate time per task. Rough percentages are fine. You want to know where your hours go.
3. Score each task. Mark it “AI can do it,” “AI can help,” or “AI can’t touch it yet.” Be honest. Test a few with an AI assistant your employer approves.
4. Find your shrinking core. If more than half your time sits in “AI can do it,” your role is likely to change significantly. That is your signal to move toward the other columns.
5. Pick 2 growth tasks. Choose tasks in the “AI can help” or “can’t touch it” columns that you want to own more of, such as client relationships, reviewing AI output, training others or solving unusual problems.

How workers can adapt

The WEF’s 2025 report found employers expect about 39% of workers’ core skills to change by 2030, and named skills gaps as the biggest barrier to business transformation. The practical response is not to become an AI engineer. It is to get good at working with AI in your own field.

• Learn the tools your industry uses. That might be an AI assistant inside your office suite, your CRM, your design tool or your code editor. Our AI skills guide for non-technical professionals has a 30-day practice plan.
• Become the checker. AI makes confident mistakes. People who can spot errors, verify facts and catch bias are in demand in every AI-heavy workflow.
• Move closer to decisions and customers. Tasks tied to accountability, trust and relationships are the hardest to hand off.
• Document your AI wins. When you use AI to save time or improve quality, write it down with numbers. It is great material for reviews and job interviews.
• Watch adjacent roles. New work is appearing around AI itself: operations, governance, quality review and agent supervision. See our guide to AI agents in 2026 for where that is heading.

For a broader view of which skills employers are hiring for, read tech careers in 2026: skills in demand.

Where to start this week

If you only have a little time, do 3 things:

1. Run the 5-step task map above on your current role.
2. Pick one weekly task in the “AI can help” column and redo it with an approved AI tool, then compare time and quality.
3. Look up your occupation in the BLS Occupational Outlook Handbook and its new AI exposure categories, and note what BLS says about your field’s 10-year outlook.

If your role sits in a high-exposure, declining group such as office support, start building a bridge skill now, while you are still employed and can practice on real work.

Conclusion

AI is changing almost every job a little and a small number of jobs a lot. The roles changing fastest lean on routine communication, data handling and first drafts. The roles holding up best combine judgment, relationships and physical skill. Official projections show real but modest declines in some office jobs, not mass unemployment, and BLS itself says the uncertainty is high. The workers who come out ahead will be the ones who treat AI as a tool to manage, not a rival to fear. For quick definitions of terms like automation and augmentation, see our glossary.

Sources

• BLS Monthly Labor Review: Incorporating AI impacts in BLS employment projections (Feb 2025) — 2023–33 occupational case studies and projected changes
• BLS Employment Projections 2025–2035 news release — total growth, office support decline, fastest-growing occupations
• BLS AI exposure categories — 4 exposure tiers, methodology and caveats
• BLS factsheet: AI impacts in Employment Projections — how BLS incorporates AI and its uncertainty
• World Economic Forum: Future of Jobs Report 2025 — skills change, fastest-declining roles, employer survey
• Anthropic Economic Index (Feb 2025) — task-level AI usage, augmentation versus automation