VeltrixVeltrix.
← All articles
13 / 62March 16, 2026

Will AI Take My Job? The Data on Which Roles Are Most and Least at Risk (2026)

Which jobs AI is already replacing, the 5 risk factors that matter, and the skills that make you AI-proof in 2026. Based on real employment data.

AI won't take your entire job. But it's already taking large chunks of tasks that currently make up your job — and that distinction matters more than people realise.

The World Economic Forum's Future of Jobs Report 2025 found that 41% of employers plan to reduce headcount in roles where AI can automate tasks by 2030. WEF That's not a distant future scenario. Companies like Klarna publicly replaced 700 customer service agents with an AI system that handles 2.3 million conversations a year. KLARMA IBM's CEO Arvind Krishna announced a hiring freeze on roles he expects AI to replace within five years, covering roughly 7,800 positions. IBM

But the same WEF report projected 170 million new jobs created by AI-driven sectors by 2030. The net isn't zero. It's a reshuffling — and the outcome for any individual worker depends almost entirely on how much of their work falls into the "routine and text-based" category versus the "contextual, physical, or relationship-intensive" category.

41%

of employers plan to reduce headcount in AI-automatable roles by 2030 WEF

170M

new jobs projected in AI-driven sectors by 2030 WEF

60%

of current jobs have at least 30% of tasks automatable with existing AI MCK

5%

of jobs are fully automatable with today's technology MCK

The key distinction

The risk isn't "will AI replace my job title" — it's "what percentage of my daily tasks involve routine text processing, data entry, or pattern matching?" That percentage is what's at risk. Your job is a bundle of tasks. Some will be automated. The question is which ones, and what that leaves you doing.

Oxford Economics researcher Carl Benedikt Frey identified the original framework for automation risk. Updated for generative AI, these five factors now determine how exposed any role is.

📄
Routine density

How much of the job involves repeating the same type of task? Data entry, standard report writing, basic customer queries — high routine density means high AI exposure.

🤝
Social complexity

Does success depend on trust, persuasion, negotiation, or reading a room? AI cannot replicate genuine human connection. Roles built on relationships are significantly safer.

🔧
Physical environment

Can the job be done entirely from a screen? Physical, manual, or unpredictable environment jobs are protected — for now. Plumbers aren't being replaced by language models.

🎨
Creative judgement

Not "making things" — that AI can do. But does the job require contextual creative judgement that can't be specified in advance? Strategy, taste, editorial decisions.

📊
Data access

Is the information required to do this job already digital and accessible? If everything needed to do the job exists in structured text, the role is more automatable than it looks.

⚖️
Accountability requirement

When something goes wrong, must a specific human be legally or professionally responsible? Liability requirements slow automation significantly in healthcare, law, and finance.

These aren't guesses. They're drawn from Oxford, McKinsey, and MIT research on actual task composition data, updated with observed AI capabilities in 2025-2026.

High risk — significant AI exposure
Data entry clerk94%
Insurance underwriter99%
Telemarketer99%
Loan officer98%
Paralegal / legal secretary94%
Basic customer service rep91%
Bookkeeper / accounting clerk97%
Travel agent97%
Low risk — genuine human advantage
Mental health counsellor0.3%
Nurse / midwife0.9%
Social worker1.4%
Plumber / electrician0.4%
Surgeon0.4%
Primary school teacher1.0%
Creative director3.8%
AI/ML engineer1.6%

The automation probability figures above come from the original Frey & Osborne methodology applied to GPT-4-level capabilities. FREY The pattern is consistent: if your job is primarily text-in, text-out with structured rules, the probability is high. If your job requires physical presence, genuine trust relationships, or contextual judgement that can't be codified, the probability is low.

One nuance worth flagging: the percentages above describe tasks within roles, not whole job elimination. A paralegal spending 80% of their time on document review has a different risk profile than one spending 80% on client relationship management and court preparation. Same job title, very different exposure.

Some of this isn't theoretical. Here's documented displacement in the sectors where AI moved fastest.

Customer serviceKlarna, BT, Vodafone
−23%
Media & journalismCNET, BuzzFeed, Sports Illustrated
−26%
Legal document reviewBigLaw firms, 2024-25
−18%
Coding (junior roles)GitHub Copilot era, 2024-25
−14%
Graphic design (stock/template)Canva AI, Adobe Firefly era
−11%
Data science (entry-level)AI-assisted analysis tools
−8%

These aren't uniform across companies or geographies — they're directional indicators from observable hiring data. The sectors seeing the most acute displacement share two traits: high volume of text-based repetitive work, and relatively standardised decision criteria that can be encoded.

The people doing best right now aren't the ones in "safe" jobs. They're the ones who've figured out how to make AI work for them rather than against them.

The single most effective thing you can do: become the person on your team who understands AI well enough to deploy it. Not to code AI — to use it, direct it, and evaluate its output. The WEF report flagged that "AI and big data" skills are the fastest-growing in-demand capability across virtually every industry through 2030. WEF

Second: audit your own tasks. For one week, track what you actually spend time on and categorise it: routine/text-based vs contextual/relationship-based. If more than 40% falls in the first category, that's where your risk is concentrated — and also where your productivity gains are waiting.

Third: stop treating AI tools as curiosities and start using them daily. The research on this is clear. A 2023 MIT study on GitHub Copilot found that experienced developers using AI assistance completed tasks 55.8% faster. MIT That productivity gap between AI-native workers and non-users compounds over time. The people at most risk aren't in high-risk roles — they're the people in any role who decide this isn't relevant to them.

The actual question to ask yourself

Not "will AI take my job?" but "am I using AI to do my job better than the version of me from 6 months ago?" The former is something that happens to you. The latter is something you control.

The workers who thrive through this transition will be the ones who treat AI fluency as a professional skill — the same way previous generations had to develop computer literacy, internet literacy, and spreadsheet skills to stay competitive.

The questions people actually search for. Direct answers.

Will AI replace software engineers?

Not wholesale, but junior engineering roles are contracting. GitHub Copilot, Cursor, and Claude handle a growing share of routine coding tasks. Senior engineers who can architect systems, review AI output critically, and understand business context are in higher demand than ever. The bottleneck is shifting from writing code to directing AI that writes code. See our AI coding tool comparison →

Will AI replace writers and journalists?

AI has already replaced the low end: SEO content farms, boilerplate press releases, standardised financial reporting. But quality journalism — the kind that requires source relationships, judgement, accountability, and a distinctive voice — remains human. CNET, Sports Illustrated, and others who deployed AI writing at scale faced significant backlash and quality problems. The lesson: AI produces volume; humans produce authority.

Will AI replace doctors?

AI is already outperforming radiologists at specific imaging tasks (detecting breast cancer from mammograms, diabetic retinopathy from retinal scans). But the clinical judgment, patient relationships, and accountability required in most medical practice aren't replicable. The likely outcome: AI-augmented doctors who can handle larger patient volumes with AI handling diagnostic support, documentation, and triage screening.

Will AI replace teachers?

The tasks will shift, not disappear. AI is already handling differentiated practice, tutoring at scale (Khan Academy's Khanmigo), and automated marking of structured assignments. What AI can't do: inspire, motivate, model human character, manage a room full of children. Teaching is one of the lower-risk professions, but teachers who use AI tools effectively will have a real advantage over those who don't.

Will AI replace accountants?

Bookkeeping, data entry, standard tax preparation — mostly automated already. The Big Four accounting firms are investing heavily in AI for audit and compliance tasks. But strategic financial advice, complex tax planning, and client relationships are safe. PwC expects AI to augment rather than replace, but their internal headcount plans tell a different story for entry-level roles. If you're in accounting, the credential that matters in 2026 isn't CPA alone — it's CPA plus demonstrated AI fluency.

Sources
WEF
World Economic Forum — Future of Jobs Report 2025weforum.org
MCK
McKinsey Global Institute — The Economic Potential of Generative AI, 2023mckinsey.com
KLARNA
Klarna — AI Assistant Results, 2024klarna.com
IBM
Bloomberg — IBM CEO AI Hiring Freeze, 2023bloomberg.com
FREY
Frey & Osborne — The Future of Employment, Oxford, 2013 (updated)oxfordmartin.ox.ac.uk
MIT
MIT — The Impact of AI on Developer Productivity, 2023papers.ssrn.com
The jobs question is the wrong question.
The right question is what you do about it.

The transition from "AI as a curiosity" to "AI as infrastructure" has already happened for forward-looking companies and workers. The 41% of employers reducing headcount in automatable roles aren't waiting for 2030. And the 170 million new jobs aren't going to workers who ignored the shift.

The data points in one direction: AI fluency is the new baseline professional skill, across every industry, at every level. The question isn't whether your job is safe. It's whether you'll be the person on your team who knows how to direct the AI that's replacing the old version of everyone's role — or the person who didn't bother learning.

Stay ahead of the curve

AI intelligence,
weekly.

Every week: the AI developments that matter, the tools worth trying, and the data behind the headlines. No hype. No filler.

Subscribe free →

Veltrix Collective · Data synthesis from WEF, McKinsey, Oxford, MIT, Klarna, IBM. Automation probability figures based on Frey & Osborne methodology applied to current AI capabilities. Job displacement data from observed hiring trends, not modelled projections. Published April 2026.

Written by Luke Madden, founder of Veltrix Collective. Data synthesis and analysis by Vel.