This is no longer a prediction. In 2025, AI job replacement moved from whitepaper projections to payroll decisions at real companies with real employees.
Klarna: 700 customer service workers replaced by AI in 2024. Duolingo, UPS, and Cisco: AI cited explicitly in headcount reductions. IBM: 7,800 positions flagged for AI replacement. In the US alone, 55,000+ layoffs in 2025 were directly attributed to artificial intelligence — and that's what companies disclosed publicly.
The question "will AI take jobs?" has been answered. The question that matters now is: which jobs are being replaced first, which are next, and is yours in either group?
Here's what the data shows.
How Many Jobs Is AI Actually Replacing?
Before the specific occupations, you need a calibrated sense of scale — because both "AI will replace everything" and "AI job fears are overblown" are wrong.
What the research says in 2026:
- MIT (November 2025): AI can now perform the equivalent work of 11.7% of the US labor force, with theoretical wage savings of $1.2 trillion. This is current capability — not a forecast.
- Goldman Sachs (2023, updated 2025): 300 million jobs globally could be partially automated by AI. Their model distinguishes "partially automated" (AI handles some tasks) from "fully replaced" (AI handles the whole role) — a distinction most headlines ignore.
- Karpathy LLM Exposure Index (2024): Scores 342 occupations on how exposed each is to large language model capabilities. The distribution is not normal — it's bimodal. Knowledge work in information-processing roles is heavily exposed. Physical work and high-accountability professional roles are not.
- Anthropic Economic Index (2024): Based on actual Claude usage patterns across industries. The sectors where AI is most actively being used to replace rather than assist human tasks: software development, customer support, content creation, and financial data processing.
The key distinction that most coverage misses: AI is replacing tasks faster than jobs. A financial analyst's job is 40% data processing, 30% synthesis, 20% communication, and 10% relationship management. AI is replacing the 40%. That's not a full job replacement — yet. But it means fewer analysts are needed to do the same work.
The gap between "AI replacing tasks" and "AI replacing people" is closing in some fields faster than others.
The 4 Patterns Behind AI Job Displacement
After reviewing the data across 1,000+ occupations, four patterns consistently determine which jobs are being replaced by AI first. These aren't factors from a 2013 automation study — they're based on how LLMs actually operate in 2024–2026.
Pattern 1: The work is primarily text-in, text-out
Large language models are — at their core — text processors of extraordinary capability. Any job whose primary output is text (reports, summaries, code, drafts, analyses, correspondence) is at high exposure. Not because AI does it better than humans — though sometimes it does — but because AI does it faster and cheaper at scale.
Most exposed: Financial analysts, writers, coders, legal document reviewers, medical transcriptionists, customer service reps handling written support.
Pattern 2: The task requires no physical presence
AI can't go anywhere. If the core of a job can be done from a terminal — processing information, generating outputs, analyzing patterns — artificial intelligence can do it without a body, a commute, or a salary. Jobs that exist only on screens are at maximum exposure.
Most exposed: Data entry, software development (routine work), content moderation, bookkeeping, research assistance, scheduling.
Pattern 3: Quality can be measured against a correct answer
AI performs best when there's a ground truth to optimize against. Code either compiles or it doesn't. Tax calculations are either right or wrong. Customer queries either get resolved or don't. When task quality is objectively measurable, AI can be evaluated, fine-tuned, and deployed at scale. When quality depends on subjective human judgment (a great sales pitch, a therapeutic breakthrough, a groundbreaking design), AI replacement is much harder.
Most exposed: Tax preparation, compliance checking, code review, data quality, medical transcription.
Pattern 4: The work is high volume, low variance
A law firm that reviews 10,000 similar contracts per year is a much easier AI replacement target than one that handles 50 bespoke transactions requiring original judgment. The pattern applies across sectors: high-volume, templated work with clear inputs and outputs is being automated first. Low-volume, novel, high-stakes work is last.
Most exposed: Document processing at scale, customer support queues, routine code maintenance, standardized report generation.
The Jobs AI Is Replacing Right Now
These occupations are experiencing active, measurable displacement — not future risk, but current reduction.
Tier 1: Already Happening at Scale (AI Risk Score 85–95/100)
Financial Analysts and Data Analysts — AI Risk Score: 95/100
The clearest case. AI has now matched or exceeded human-level performance on financial report generation, earnings analysis, and quantitative modeling — the core tasks of entry-to-mid-level financial analysts. Major banks including Goldman Sachs, JPMorgan, and BlackRock have deployed AI systems that generate first-draft research reports. The number of junior analysts needed is declining. Senior analysts who interpret, synthesize, and communicate with clients are safer — but the pipeline is narrowing.
Computer Programmers (Routine Work) — AI Risk Score: 93/100
GitHub Copilot, Cursor, and Claude Code are not just productivity tools — they're labor substitutes for routine code generation. Entry-level developer hiring at major tech companies contracted sharply in 2024–2025. The pattern: AI handles boilerplate, CRUD, test generation, and documentation. Human developers doing this work face real displacement pressure. Engineers who architect systems, manage AI output, and make technical judgment calls are in a different risk category.
Medical Transcriptionists — AI Risk Score: 92/100
This is the most complete AI replacement story already playing out. AI transcription tools like Nuance DAX and ambient clinical documentation software are being deployed at health systems nationwide. The occupation is contracting in real-time. This isn't a warning — it's an obituary for a profession that existed for decades.
Customer Service Representatives (Written/Chat Support) — AI Risk Score: 74/100
Klarna's much-publicized case is not an outlier — it's a preview. AI handles 2.3 million customer conversations per month for Klarna, doing the work that previously required 700 employees. The economics are unambiguous: AI support costs a fraction of human support for the large majority of standard queries. What AI can't replace yet: complex escalations, emotionally charged situations, and customers who specifically demand human interaction. These tasks remain — but the job category is shrinking.
Tier 2: Fast-Moving Displacement (AI Risk Score 70–84/100)
Writers and Content Creators (Volume Work) — AI Risk Score: 70/100
The category is bifurcating sharply. High-volume, SEO-driven, templated content is being heavily automated — content farms and many in-house content roles have already shrunk. Original journalism, distinctive brand voice, investigative work, and high-judgment editorial roles are less exposed. The middle — competent, reliable writers producing solid-but-unremarkable work — is being squeezed hardest.
Accountants and Bookkeepers — AI Risk Score: 74/100
AI is transforming accounting from the bottom up. Routine bookkeeping, bank reconciliation, basic tax preparation, and compliance reporting are all automatable with current tools. CPA firms are already deploying AI for first-pass work that junior accountants used to handle. The safe end of accounting: advisory work, complex tax strategy, M&A due diligence, and client relationship management. The at-risk end: entry-level processing and compliance work that once provided the on-ramp to the profession.
Paralegals and Legal Document Reviewers — AI Risk Score: 68/100
Contract review, e-discovery, and legal research are the three paralegal functions most exposed to AI replacement. Tools like Harvey and Spellbook can review contracts at scale, flag risks, and summarize clauses with accuracy that matches or exceeds junior paralegals. Law firms are restructuring staffing models. Junior and mid-level document review work is the clearest displacement story in legal.
Marketing Managers (Campaign/Performance) — AI Risk Score: 88/100
Performance marketing — ad copy generation, A/B test management, campaign optimization, audience targeting — is being heavily automated. AI tools now handle tasks that required a team of specialists five years ago. Brand strategy, creative direction, and senior marketing leadership remain more protected, but the execution layer of marketing is contracting.
Tier 3: Emerging Pressure (AI Risk Score 55–69/100)
Project Managers (Admin-Heavy Roles) — AI Risk Score: 60/100
Status reporting, timeline tracking, meeting summaries, resource allocation optimization — these are the administrative core of many project management roles, and AI handles them well. Senior PMs whose value comes from stakeholder relationships, conflict resolution, and strategic judgment are in a better position. Administrative PMs are not.
Software Quality Assurance Testers — AI Risk Score: 72/100
AI-driven test generation and automated QA pipelines are reducing the need for manual test writing. The category is under active compression.
Journalists and News Writers (Commodity News) — AI Risk Score: 65/100
Wire service rewrites, earnings reports, sports scores, and weather summaries are already AI-generated at many outlets. Investigative journalism, distinctive columnists, and reporters with deep source networks are in a different risk tier.
The Data Behind AI Job Displacement
Most "AI will replace jobs" headlines still cite the Frey & Osborne 2013 Oxford paper — which predicted 47% of US jobs at risk. That paper was published before ChatGPT existed. It modeled robotic process automation and narrow AI, not large language models.
Our risk scores use three 2023–2024 sources:
Karpathy LLM Exposure Index (40% weight)
Andrej Karpathy scored 342 occupations task-by-task on LLM exposure. This is current AI capability, not a 10-year forecast.
OpenAI "GPTs are GPTs" — Science, 2024 (30% weight)
Academic research measuring what percentage of each occupation's core tasks LLMs can perform at human parity. Peer-reviewed methodology using US BLS occupation standards.
Anthropic Economic Index, 2024 (30% weight)
Based on actual Claude deployment data — where AI is actually being used to replace or assist human work by industry. This is revealed behavior, not a model.
Composite score: raw = Karpathy×0.4 + OpenAI×0.3 + Anthropic×0.3, clamped to 40–95.
The result is the most current multi-source AI job displacement methodology available — and it produces meaningfully different rankings than outdated studies.
Which Jobs Are Safe from AI Replacement?
If your occupation is in the at-risk categories, understanding the protected end of the spectrum matters. The full breakdown is in our jobs AI can't replace guide, but the headline findings:
The jobs AI is not replacing in 2026:
- Hands-on healthcare — nurses (40/100), surgeons (45/100), physical therapists (46/100). Physical presence + accountability.
- Mental health and social services — therapists (38/100), social workers (38/100), clergy (38/100). Human relationship is the work.
- Skilled trades — electricians (38/100), plumbers (41/100), roofers (36/100). Physical complexity in uncontrolled environments.
- Early education — elementary teachers (38/100). Children need human mentors, not AI instructors.
- Senior leadership — executives who bear accountability for consequential decisions.
The pattern: physical presence, human accountability, or emotional relationship creates structural protection. Information processing at scale does not.
See the full list of AI-resistant careers →
Is Your Job Being Replaced?
Your job title doesn't tell you enough. A "software developer" who architects systems is different from one writing CRUD endpoints. A "financial analyst" with deep client relationships is different from one who generates standard reports.
The question is which tasks you spend your time on — and how many of those map to Pattern 1 through 4 above.
Take the free 3-minute AI risk assessment →
Enter your job title, answer 4 questions about your actual work, and get a personalized risk score. You'll see exactly which of your skills are most exposed to AI replacement — and which ones are protecting you.
It's free. The data comes from the same 2023–2024 research behind this guide.
If Your Job Is at Risk: What Actually Works
The professionals who navigate AI job displacement successfully are not the ones who panic — or the ones who ignore it. They're the ones who made one or two deliberate moves before it became urgent.
Move 1: Stop competing with AI on its strengths
AI is fast, cheap, and tireless at text processing. If your job is 80% generating structured text from structured data, you're in direct competition with something that costs $0.01 per 1,000 words. Stop competing there. Find the 20% that requires judgment, relationships, or accountability — and make that your expansion zone.
Move 2: Become the person who directs AI
Every field is splitting into two tracks: those using AI to multiply their output by 5x, and those being replaced by those people. The prompt engineer, the AI output reviewer, the professional who evaluates and refines AI work — these roles are being created by AI displacement, not eliminated by it.
Move 3: Move up the accountability chain
The single best predictor of AI displacement resistance: who bears legal and professional accountability for the outcome. Junior work with no accountability is at maximum risk. Senior work where someone's license, reputation, or freedom is on the line is the most protected. Deliberately move toward the parts of your field where you own the outcome.
Move 4: Read the concrete plan
How to Future-Proof Your Career Against AI → is a 90-day practical guide for workers in mid-to-high risk occupations. Not abstract advice — specific skills, specific moves, specific timelines.
Frequently Asked Questions
How many jobs will AI replace by 2030?
Estimates vary significantly depending on methodology. Goldman Sachs projects 300 million jobs globally will be partially automated by AI. MIT's 2025 research found AI can currently perform the equivalent work of 11.7% of the US labor force. The World Economic Forum's 2025 Future of Jobs report estimated 85 million jobs would be displaced by automation by 2025 — while 97 million new AI-adjacent roles would be created. The net effect depends heavily on transition speed and policy response. What is measurable now: 55,000+ US layoffs in 2025 were directly attributed to AI.
What jobs are being replaced by AI right now?
The occupations facing real-time displacement in 2025–2026: medical transcriptionists (AI transcription is near-complete), junior financial analysts (AI report generation), routine customer service (AI chat at scale), entry-level software developers (AI code generation), legal document reviewers (AI contract analysis), and high-volume content writers. These aren't forecasts — they're observable in hiring data and company disclosures.
Is AI job replacement accelerating in 2026?
Yes, and the acceleration is driven by three factors converging: AI capability has crossed the threshold for "good enough" in more task categories, AI costs have dropped dramatically making deployment economically obvious, and enough early adopters have demonstrated ROI that risk-averse organizations are now following. The 2025–2026 window is when AI job displacement moved from exceptional to mainstream.
Which jobs are NOT being replaced by AI?
The jobs with the lowest AI replacement risk are those requiring physical presence (skilled trades, hands-on healthcare), human emotional relationship (therapists, teachers, clergy), or high-stakes accountability (surgeons, lawyers, senior executives). See the full breakdown: Jobs AI Can't Replace →
Will AI replace entry-level jobs first?
Yes — this is one of the clearest patterns in the data. Entry-level roles that primarily involve producing structured outputs under supervision (junior analysts, associate developers, content assistants) are being replaced faster than senior roles. This creates a "missing rung" problem: the entry-level jobs that used to be the path to senior expertise are disappearing, while senior roles remain safer. This is one of the most underreported dimensions of AI job displacement.
How do I find out if AI is replacing my job?
The fastest way: take the free assessment at AIProofMe. You'll get a risk score for your specific occupation and task mix, based on the same 2023–2024 research data (Karpathy, OpenAI, Anthropic) behind this analysis. Alternatively, browse occupation scores at /will-ai-replace — we've scored 1,000+ roles.
Is AI job loss overstated?
The media coverage is often poorly calibrated — alternating between "AI will replace everything" and "the fears are overblown." What the careful research shows: AI is displacing tasks faster than jobs, the impact is highly uneven across occupations, and there's a real difference between "augmentation" (AI makes this person more productive) and "replacement" (AI makes this person unnecessary). In some categories — medical transcription, entry-level financial analysis, routine code generation — replacement is real and measurable. In others, augmentation is the dominant story. The honest answer is: it's neither as apocalyptic nor as benign as the extremes suggest.
Explore AI Risk by Occupation and Industry
See how AI job displacement is playing out in your sector:
- Will AI replace accountants? — Score 74/100. The processing tier is at risk; the advisory tier is safer.
- Will AI replace lawyers? — Score 63/100. Document review is exposed; courtroom and advisory work is not.
- Will AI replace software developers? — Score 93/100 for routine work. Architects and senior engineers score significantly lower.
- Will AI replace doctors? — Score 45/100. Diagnostic AI assists; clinical accountability remains human.
- Will AI replace teachers? — Score 38/100. One of the most protected major employment sectors.
- AI risk in finance — Score 95/100. The highest-risk individual occupation by our data.
- AI risk in technology — Sector average 69/100. High variance — safe at the senior end, exposed at junior.
Methodology: AI risk scores use a weighted composite of Karpathy LLM Exposure Index (40%), OpenAI "GPTs are GPTs" Science paper (30%), and Anthropic Economic Index (30%), clamped to 40–95. All data from 2023–2024. Full methodology →