Will AI Take Your Job? 20 Careers AI is Already Reshaping

From radiologists and lawyers to coders and consultants, AI is now doing work that once took years of training. We dug into the latest research from Stanford, Microsoft, the World Economic Forum, the ILO and others to see which careers are most exposed, and what that really means.

For generations the advice was simple: study medicine, law, engineering or finance, earn the credential, and you're set for life.

That promise is now being tested. AI systems are catching cancers radiologists miss, writing production code, winning gold at the International Mathematical Olympiad and drafting legal briefs. And for the first time, the effects are showing up in payroll data.

AI is hitting entry-level jobs first

In August 2026, Stanford's Digital Economy Lab updated its landmark study, Canaries in the Coal Mine?, using ADP payroll records covering millions of American workers through June 2026. The headline finding:

• 19%: employment among 22- to 25-year-olds in the most AI-exposed occupations is now about 19% below where it would be had it kept pace with their peers in less-exposed jobs. That gap was 15% in the first version of the study, published in August 2025, so it is widening.
• -11% vs +10%: in raw numbers, employment of young workers in the most exposed jobs fell about 11% between November 2022 and June 2026, while young workers in the least exposed jobs grew about 10%.
• No gap for veterans: experienced workers in the same exposed occupations show no comparable decline.
• Hiring, not firing: the gap is driven mainly by companies hiring fewer young people, not by layoffs, and it is concentrated in jobs where AI automates tasks rather than assists with them. Where AI complements workers, employment is flat or rising.

The authors are careful. They describe these as descriptive patterns, not proof that AI caused the gap, and they still see no sign of economy-wide job loss. Other researchers using national survey data have found smaller effects.

But the signal is hard to ignore. AI may not be eliminating careers. It may be quietly removing the first rung of the ladder.

Which jobs are most at risk from AI?

There is no single "AI risk" league table, but four major studies point in the same direction: language-heavy, screen-based knowledge work is most exposed, and hands-on physical work is least.

Microsoft: 200,000 real conversations

Microsoft researchers analysed 200,000 anonymised Bing Copilot conversations, mapped them to the US government's O*NET database of work activities, and gave every occupation an "AI applicability" score based on how often, how successfully and how broadly AI is used for that job's tasks. Here is where careers on our list land among the top 40 most exposed:

Microsoft Rank	Occupation	AI applicability score
1	Interpreters and translators	0.49
2	Historians	0.48
5	Writers and authors	0.45
6	Customer service representatives	0.44
9	Ticket agents and travel clerks	0.41
10	Broadcast announcers and radio DJs	0.41
13	Telemarketers	0.40
16	News analysts, reporters, and journalists	0.39
17	Mathematicians	0.39
19	Proofreaders and copy markers	0.38
21	Editors	0.37
29	Data scientists	0.36
33	Web developers	0.35
34	Management analysts	0.35
37	Market research analysts	0.35

At the other end, jobs that seem protected from AI disruption are phlebotomists, nursing assistants, hazardous materials removal workers, dredge operators and embalmers. Microsoft also found that jobs requiring a bachelor's degree score only slightly higher on average than those that don't, a sign that AI's reach now extends to graduate jobs, not just routine ones.

What employers plan to cut

The World Economic Forum's Future of Jobs Report 2025 surveyed more than 1,000 employers. It projects 170 million jobs created and 92 million displaced by 2030, a net gain of 78 million, but 41% of companies plan to shrink their workforce because of AI. The fastest-declining roles are led by clerical work (postal clerks, bank tellers, data entry clerks, administrative assistants), with accountants and auditors ranked 7th. For the first time, graphic designers and legal secretaries entered the top 10, a sign that generative AI is reaching creative and professional work.

The global picture from the ILO

The International Labour Organization's 2025 index, built on nearly 30,000 tasks, finds that one in four workers worldwide is in a job with some exposure to generative AI. That rises to 34% in high-income countries. Only 3.3% of global jobs fall in the highest-exposure category, and women are more exposed than men: in rich countries, 9.6% of female workers are in the top exposure band versus 3.5% of men.

The reason is the jobs women hold. Women are overrepresented in clerical, administrative and business support roles, such as secretaries, administrative assistants, data entry clerks and bookkeepers. As a result, female-dominated occupations are almost twice as likely to be exposed to generative AI as male-dominated ones: 29% versus 16%.

Anthropic: who is actually using AI

Anthropic's Economic Index, drawn from millions of Claude conversations, shows how uneven adoption is. Computer and mathematical jobs account for 37% of usage despite making up just 3.4% of US workers. Around 36% of occupations use AI for at least a quarter of their tasks, but only about 4% use it for three-quarters or more, a reminder that most jobs are being reshaped task by task, not wiped out wholesale.

20 careers AI is already reshaping

This is not a list of jobs about to disappear. It is a research-backed watchlist of 20 careers where AI is already doing important parts of the work. We chose them because they score high on AI exposure in the Microsoft and ILO data, show real evidence of change in hiring, pay or output, and together cover a range of sectors. For each, we've pulled the strongest, most concrete data point we could find.

1. Radiologists

The data: In Sweden's MASAI trial of more than 105,000 women, AI-supported screening found 29% more cancers and cut radiologists' reading workload by 44%, with no measurable rise in false alarms. Final results, published in The Lancet in January 2026, showed 12% fewer cancers slipping through between screenings.

The twist: In 2016, AI pioneer Geoffrey Hinton said we should stop training radiologists. A decade later there is a global shortage. Imaging volumes keep climbing, and average US radiologist pay hit about $571,000 in 2025, up 9% in a year. AI is absorbing the growth, not the jobs.

The question: Will radiologists become supervisors of AI that handles routine screens, while they focus on complex cases and procedures?

2. Software engineers

The data: On SWE-bench Verified, a test built from real GitHub bug reports, top AI models went from solving under half of issues in late 2023 to about 80% by early 2026. Coding is also AI's single biggest use case: computer and maths jobs account for 37% of all Claude conversations. But passing a test isn't the same as shipping code. In March 2026, the AI research group METR found that roughly half of test-passing AI pull requests would not be merged into a real codebase by its maintainers, mostly over code quality and problems the tests missed. Strong human engineers are still needed to judge what is good enough to ship.

The squeeze: Stanford found employment of software developers aged 22 to 25 fell nearly 20% from its late-2022 peak, while developers aged 35 to 49 grew. On freelance platforms, demand for software and web development fell about 21% within eight months of ChatGPT's launch. Yet the BLS still projects overall software developer employment to grow over the decade.

The question: If one senior engineer with AI can do the work of a small team, where do the next senior engineers come from?

3. Lawyers

The data: Professionals in Thomson Reuters' 2025 Future of Professionals survey expect AI to save them 5 hours a week, about 240 hours a year. The WEF lists legal secretaries among the 10 fastest-declining jobs for the first time, and the BLS projects legal secretaries and administrative assistants to shrink about 6% by 2034.

What's exposed: document review, legal research, contract drafting and first-pass summaries, the work junior associates have always cut their teeth on. Accountability, advocacy, negotiation and client trust stay human.

The question: If AI handles the grunt work, how many junior associates will firms hire, and how will they learn? And in 20 years, who will be ready to replace today's partners and senior lawyers?

4. Mathematicians

The data: In July 2025, Google DeepMind's Gemini Deep Think scored 35 out of 42 at the International Mathematical Olympiad, solving five of six problems for an officially certified gold medal. OpenAI reported the same score with an experimental model. A year earlier, the best AI managed silver. Microsoft ranks mathematicians #17 for AI applicability.

The limit, and how fast it's moving: Olympiad problems have known answers. Research mathematics means inventing the questions, building new theory and proving things no one has proved before. Yet in May 2026, an OpenAI model disproved Paul Erdős's 1946 "unit distance" conjecture, an 80-year-old problem most mathematicians believed was true, producing 125 pages of original mathematics from a single prompt. Mathematicians still had to verify and refine it, and its edge was persistence more than new ideas: it assumed Erdős was wrong and kept applying known techniques until it found a counterexample.

The question: As AI handles more calculation and proof-checking, will mathematicians become architects of ideas rather than solvers of problems?

5. Financial analysts

The data: Microsoft ranks personal financial advisors #30 and brokerage clerks #11 for AI applicability.

What's exposed: reading filings, summarising earnings calls, building first-draft models and writing research notes, the core of a junior analyst's week. Judgment on markets, incentives and risk still sets the best analysts apart.

The question: Will banks and funds keep hiring large analyst classes when AI drafts the first version of the research?

6. Management consultants and analysts

The data: In a landmark Harvard Business School study of 758 BCG consultants, those using GPT-4 completed 12% more tasks, 25% faster, with 40% higher quality. The weakest performers gained the most. But on a task just beyond AI's abilities, consultants using it were 19 percentage points less likely to get the right answer. Microsoft ranks management analysts #34.

The squeeze: Market research, competitor scans and slide-building are exactly what AI does well, and exactly how junior consultants learn the trade: by doing the research and building the decks.

The question: What happens to the consulting pyramid when the bottom layer can be automated?

7. Pathologists

The data: Paige Prostate, the first AI pathology tool authorised by the FDA, raised pathologists' cancer detection from about 89% to 97% in its pivotal study, cutting missed diagnoses by 70%. With its help, generalist pathologists performed as well as prostate specialists.

The limit: Clinical use demands rigorous validation, and diagnoses still carry a pathologist's signature and accountability.

The question: If AI lets a general pathologist read slides like a subspecialist, does the subspecialist become rarer, or does their role move to the hardest cases and to checking the AI? And does that mean more access to care, or fewer pathologists?

8. Accountants and auditors

The data: The WEF ranks accountants and auditors 7th among the world's fastest-declining jobs to 2030, with bookkeeping and payroll clerks 6th. In the UK, the Big Four cut graduate and school-leaver intake from about 6,500 in 2023 to 5,400 in 2025. KPMG alone cut its intake by 29%.

The paradox: In the US, firms still report shortages of qualified CPAs. The work isn't disappearing so much as moving up: first-year staff who once spent weeks reconciling ledgers now review what AI produces in hours.

The question: If AI does the work trainees learn on, can we be sure the profession can still train its next generation effectively?

9. Architects

The data: The share of UK architecture practices using AI jumped from 41% in 2024 to 59% in 2025, according to the Royal Institute of British Architects. Only 18% of architects expect job losses as a direct result, but 35% see AI as a threat to the profession and 44% worry it will let unqualified people produce building designs.

What's exposed: early concept generation, visualisation, documentation and option testing. Regulations, safety, site realities and client relationships stay firmly human.

The question: Will firms need as many junior designers and drafters when AI can produce 50 concept options overnight?

10. Translators and interpreters

The data: This is the most exposed job in Microsoft's entire ranking: #1 of every occupation it scored, with AI successfully used on 98% of its core work activities. In a 2024 Society of Authors survey, 36% of translators said they had already lost work to generative AI and 43% said their income had fallen.

What's left: Literary translation, legal and medical interpreting, and culturally sensitive work still need humans, and increasingly the job is editing machine output rather than translating from scratch.

The question: Is routine translation becoming a near-free commodity, with humans left to the premium niche?

11. Customer service reps and telemarketers

The data: Microsoft ranks customer service reps #6 and telemarketers #13. The BLS projects customer service jobs to shrink 5% by 2034, about 154,000 fewer jobs, and telemarketing to fall about 22%, one of the steepest declines of any occupation. In 2024 Klarna said its AI assistant was handling two-thirds of customer chats, the equivalent work of 700 full-time agents.

The reality check: By 2025, Klarna's CEO admitted the cost-cutting push had gone too far and the company began recruiting human agents again so customers could always reach a person. AI handles the routine; humans handle the angry, the complex and the unusual. On outbound calls, regulation of automated calling and consumer distrust of robocalls slow things further.

The question: Will customer service become a small, highly skilled human team supervising fleets of AI agents, and how much selling will still be done by a human voice by 2030?

12. Historians

The data: Historians rank #2 on Microsoft's list, which found AI applicable to around 91% of a historian's typical tasks, from gathering sources to summarising and writing them up. AI is already doing real historical work: Google DeepMind's Aeneas, published in Nature in July 2025, restores damaged Latin inscriptions with 73% accuracy when up to ten characters are missing, and historians found its results useful as research starting points.

The pushback: A comment in the Royal Historical Society's journal argued that Microsoft's ranking rests on an oversimplified picture of what historians actually do. It is also a small profession, about 3,040 people in the US, so this is about AI changing the work, not mass job losses.

The question: If AI can read, date and summarise the archive, does the historian's value shift entirely to interpretation, deciding what the past means?

13. Writers, authors, proofreaders and copy editors

The data: Writers and authors rank #5 on Microsoft's list, proofreaders #19 and editors #21. On freelance platforms, writing jobs were hit hardest of all: postings fell about 30% within eight months of ChatGPT's launch, according to the Imperial, Harvard and DIW study. A separate study of more than 92,000 Upwork writers found monthly earnings dropped about 5% within months. AI now catches spelling, grammar and style-guide errors in seconds.

What's left: original ideas, a distinctive voice, fact-checking, structural editing and judgment on tone, the parts of writing and editing AI still gets wrong.

The question: Will writing and editing shift from producing and polishing words to safeguarding originality, accuracy and credibility?

14. Administrative assistants

The data: Administrative assistants and executive secretaries are among the WEF's top fastest-declining roles. The BLS expects office and administrative support to lose about 752,000 jobs by 2034, the largest decline of any major occupational group.

What's left: Experienced executive assistants bring discretion, judgment and relationship management that AI scheduling tools can't.

The question: As every professional gets an AI assistant, what happens to the human ones?

15. Data scientists

The data: Here's the paradox. Microsoft ranks data scientists #29 for AI exposure, yet the BLS projects data scientist jobs to grow 33.5% between 2024 and 2034, one of the fastest rates of any occupation, driven largely by AI adoption itself.

What's exposed: writing analysis code, cleaning data, building charts and explaining results. What isn't: choosing the right method, spotting bad data, and telling correlation from causation.

The question: Will demand keep growing for senior data scientists who direct AI, while junior analyst work gets automated?

16. Graphic designers

The data: For the first time, the WEF placed graphic designers in its top 10 fastest-declining jobs. On freelance platforms, demand for graphic design and 3D modelling fell about 17% after image generators like Midjourney and DALL-E arrived, with no sign of recovery, according to researchers at Imperial College, Harvard and the German Institute for Economic Research.

What's left: brand strategy, art direction, visual systems and the taste to know what's good.

The question: Will designers shift from making individual assets to directing whole creative systems?

17. Recruiters

The data: LinkedIn's AI Hiring Assistant, rolled out globally in 2025, lets recruiters review 62% fewer profiles, lifts message acceptance rates 69% and saves more than 4 hours per role, according to LinkedIn.

The nuance: LinkedIn says hiring overall is down about 20% since 2022, but blames higher interest rates, not AI. Meanwhile, AI screening raises serious questions about bias, fairness and privacy.

The question: If AI does the sourcing and screening, does the recruiter become a talent adviser, or redundant?

18. Journalists and reporters

The data: News analysts, reporters and journalists rank #16 on Microsoft's list. A Reuters Institute survey of 1,004 UK journalists found 56% use AI at least weekly, mostly for transcription, translation and copy-editing. Only 10% use it for first drafts. In the Reuters Institute's 2026 survey of media leaders, just 38% said they were confident about journalism's prospects.

What's left: original reporting, building sources, investigations and verification, the boots-on-the-ground work that makes journalism worth paying for.

The question: Will newsrooms automate routine content and reinvest in reporting, or simply shrink?

19. Travel agents

The data: Microsoft ranks ticket agents and travel clerks #9 and concierges #14. Yet travel agent employment has held up better than you might expect: the BLS projects little or no change over the next decade, about 61,500 jobs today.

Why: Much of the routine booking work moved online two decades ago. What's left is complex itineraries, luxury and specialist travel, group trips and sorting out disruptions, work built on relationships.

The question: Will AI travel assistants finish off DIY trip planning, leaving human advisers with the premium end?

20. Radio DJs and broadcast announcers

The data: Broadcast announcers and radio DJs rank #10 on Microsoft's list. In 2025, it emerged that "Thy", host of the four-hour weekday show Workdays with Thy on Sydney station CADA, was AI-generated, with a voice built using ElevenLabs. She had been on air for about six months before anyone found out. What gave her away wasn't her voice: she had no last name, no biography and no social media presence.

What's left: live local presence, personality, community connection and covering news as it breaks.

The question: If listeners can't tell the difference, will they still care whether the voice is human?

Exposure is not extinction

If there's one lesson from the data, it's this: being exposed to AI is not the same as being replaced by it.

Radiology is the proof. It was supposed to be the first profession AI killed. Instead it has a shortage, rising pay and AI tools that help each radiologist handle more scans. The ILO reaches the same conclusion globally: because most jobs still need human input, transformation is far more likely than redundancy. Anthropic's data agrees, with only about 4% of occupations using AI for most of their tasks.

The real question is economic, not technical. When AI makes a lawyer, coder or analyst twice as productive, does the employer serve twice as many clients, or hire half as many people? The answer will differ by industry, by country and by company, and it is being decided right now.

The evidence is also still contested. Researchers in Denmark found AI was not the main driver of falling early-career employment there, and other studies of job postings find junior and senior roles declining at similar rates. Anyone claiming certainty about AI and jobs is getting ahead of the data.

What should students study to be AI-proof?

There is no AI-proof degree. But the research points to a clear pattern in what holds its value:

• Judgment over routine: the tasks AI struggles with are those that need context, accountability and decisions under uncertainty.
• People skills: Microsoft's least-exposed jobs, from nursing assistants to phlebotomists, are hands-on and human-facing.
• AI fluency: the BCG study showed that people who used AI well got dramatically better, and people who used it carelessly got worse. Knowing when the AI is wrong is now a core professional skill.
• Deep expertise: generalists using AI can match specialists on routine work, which makes true expertise on the hard cases more valuable, not less.

The goal is not to avoid medicine, law, engineering or finance. It is to understand how those professions are changing, and to build the skills that make you the person directing the AI, not the one it replaces.

The bigger question: who trains the next generation?

The most worrying signal in the data isn't that whole professions are vanishing. It's where the pressure is landing: on the first rung of the ladder.

Stanford finds a 19% gap for young workers in AI-exposed jobs and none for experienced ones. The UK's Big Four have cut graduate intake by more than 1,000 places in two years. In Thomson Reuters' survey, almost a quarter of professionals worried that relying on AI could stunt professional development.

For generations, junior lawyers learned by reviewing documents, junior analysts by building models and junior consultants by building decks. If AI does that work, the apprenticeship model that produced every senior expert breaks.

Some employers are already rethinking it. Reuters' editor-in-chief has argued AI should free young reporters from desk work so they can get back into the field. That may be the real opportunity: not fewer entry-level jobs, but different ones, where young professionals learn judgment earlier because the grunt work is gone.

The future of work won't be defined only by which jobs AI replaces. It will be defined by how we rebuild the ladder.

FAQ: AI and your job

1. What jobs will AI replace first?

Language-heavy, screen-based knowledge work is most exposed. Microsoft's 2025 study of 200,000 Copilot conversations found interpreters and translators, historians, passenger attendants, sales representatives and writers had the highest AI applicability scores. Exposure means AI can do parts of the job, not that the job will disappear.

2. Which jobs are safest from AI?

Hands-on, physical and people-facing work. Microsoft's least-exposed occupations include phlebotomists, nursing assistants, hazardous materials removal workers, dredge operators and embalmers.

3. Will AI replace entry-level jobs?

The early evidence points that way. Stanford's August 2026 update found employment of 22- to 25-year-olds in the most AI-exposed jobs is about 19% below where it would otherwise be, driven by fewer hires rather than layoffs. Experienced workers in the same jobs show no comparable decline.

4. Will AI cause mass unemployment?

Not so far. The World Economic Forum projects 170 million jobs created and 92 million displaced by 2030, and the ILO finds only 3.3% of global jobs in the highest exposure band. Most jobs are being reshaped task by task.

5. Is any job AI-proof?

No degree is fully AI-proof. The skills that hold their value are judgment, people skills, AI fluency and deep expertise on the hardest cases.

 

Learn more about the state of AI today, by downloading our STATE OF AI 2026 REPORT here. 

---

Lucidity Insights is exploring the future of work in its "Is Your Job AI-Proof?" video series. This article is an editorial watchlist, not a forecast that the listed professions will disappear, and will be updated as new research becomes available.