Updated June 2026
The Future of Work After AI: Which Jobs Disappear and Which Ones Boom
In 2016, a Turing Award winner stood on a stage in Toronto and told an entire profession to stop training new members. Geoffrey Hinton said radiologists had five years left, maybe ten. Reading a 2026 report on what actually happened, the math runs the other direction: radiologist pay has climbed to roughly $571,000 a year, and there were more than 4,000 unfilled radiology jobs sitting open for months at a time. The most confident person in the room about AI and jobs was wrong about the one job he picked.
That should bother you more than it does. Most coverage of AI and employment still treats the question the way Hinton did in 2016 — find the job, draw the curve, announce the date. The actual research, the kind buried in PDFs from Stanford's Digital Economy Lab or PwC's billion-job-ad dataset, tells a messier story where the same occupation can show up on both the endangered list and the hiring-boom list depending on which slice of the job you're measuring. Most "jobs AI will kill" listicles never make it past slice one.
What follows draws on the World Economic Forum's 2026 jobs data, Goldman Sachs' labor economists, a Stanford study that triggered a genuine academic fight, and PwC's analysis of a billion job postings. Fifteen disappearing roles, fifteen booming ones, and the parts in between that the clean lists leave out.
- The number that should reframe this whole debate
- 15 jobs losing ground
- The entry-level argument nobody has settled
- 15 jobs gaining ground
- The Hinton problem: why predictions keep missing
- Who actually needs to act on this right now
- Verdict
- FAQ
The number everyone skips past
Start with the figure most explainer articles bury at paragraph nine. The World Economic Forum's Future of Jobs Report 2026 projects 92 million roles displaced by 2030 against 170 million newly created — a net gain of 78 million jobs globally. Net positive. That's the headline every press release leads with, and it's also the number that tells you the least about whether your specific job is fine.
Net figures average over a labor market that isn't behaving like an average. PwC's 2026 Global AI Jobs Barometer, built from more than a billion job postings, found that companies most exposed to AI are growing headcount and wages faster than companies least exposed — by a wide margin, with the top fifth of exposed firms posting 163% productivity growth. Meanwhile Goldman Sachs' research team estimates that if AI use cases were rolled out economy-wide today, only about 2.5% of US employment would be at risk — a fraction of the apocalypse headlines suggest.
Here's the number nobody puts in the explainer: PwC found the skills gap between AI-exposed and AI-sheltered jobs is widening 75% faster than it was just one year earlier. The split isn't stabilizing. It's accelerating.
So both things are true at once. The aggregate looks fine. Inside the aggregate, a two-track labor market is forming, and which track you're on depends less on your job title than on whether your specific tasks are routine, structured, and text-or-data-based — the three traits AI eats first.
15 jobs losing ground
None of these vanish overnight. What's shrinking is headcount per unit of output — fewer people doing the same volume of work, which functions the same as job loss for anyone competing for the remaining seats.
- Data entry clerks face the highest exposure of any category measured, with OCR and AI now handling ingestion, validation, and formatting end to end.
- Customer support agents handling Tier 1 tickets are being displaced fastest where AI chat systems can resolve queries from a trained knowledge base without escalation.
- Telemarketers appear on Goldman Sachs' own high-risk list, alongside the broader collapse of outbound call-center work.
- Paralegals doing document review and contract comparison are losing hours to AI tools that can flag clauses and precedent faster than a junior associate.
- Junior financial analysts who spend their days compiling reports are competing with dashboards that flag anomalies and draft summaries automatically.
- Copywriters and content assistants producing first-draft marketing copy sit in the 60–75% exposure range cited by multiple labor-market trackers.
- Bookkeepers and accounts-payable clerks are losing the reconciliation and categorization work that used to fill a full-time role.
- Proofreaders and copy editors appear by name on Goldman Sachs' high-risk occupation list.
- Credit analysts doing routine risk scoring are also named explicitly in that same Goldman Sachs analysis.
- Administrative assistants and clerical staff are the single occupation category the WEF flags as declining across every version of its jobs report since 2025.
- Retail cashiers face displacement as self-checkout and AI-driven point-of-sale systems extend further into stores.
- Translators and interpreters doing routine document or live-caption work are losing ground to real-time AI translation that keeps closing the quality gap.
- Junior software developers writing boilerplate code are affected less by elimination than by firms simply hiring fewer of them per project.
- Graphic designers producing template-based marketing assets are competing against AI image generation tools that now handle a meaningful share of that volume.
- Print production and pre-press workers continue a decline that predates generative AI but has accelerated alongside it.
Look at that list again. Almost every entry shares a trait: the job is mostly one task, repeated. Hinton's mistake with radiology was assuming a profession is the sum of its most automatable task. The jobs above are vulnerable precisely because, for many workers in these roles, that assumption happens to be closer to true.
[No relevant internal article found on peakoftrending.blogspot.com for this topic.]
The entry-level fight nobody has settled
This is the part of the AI-jobs story that has an actual ongoing dispute, not just a list of predictions. In August 2025, Stanford's Digital Economy Lab published findings that employment for workers aged 22 to 25 in the most AI-exposed occupations had declined substantially relative to older workers in the same fields, while wages stayed roughly flat — meaning the damage was landing on headcount, not pay. Software engineering and customer service led the decline. Microsoft's own 2026 labor research cited the same 16% relative drop for that age cohort in highly exposed roles. You graduate, you've done the internships, your resume looks like the one that got your older sibling a job three years ago. The postings have the same titles. The headcount behind them has quietly shrunk.
Then a second group disagreed, and not quietly. The Stanford Review — student-run, but citing Stanford's own Professor Eric Roberts — argued the timing is a coincidence stacked on a different cause: interest rate hikes choking tech hiring, the same pattern that hit computer science graduates after the dot-com crash and reversed within a couple of years once rates eased. Same campus. Same data window. Opposite conclusion about what's actually driving the collapse.
Nobody has resolved this. The Stanford Digital Economy Lab researchers call the trend a "canary in the coal mine." The Review calls AI a "convenient scapegoat" for a macroeconomic problem. Both sides are looking at the same hiring charts.
The senior talent that companies are retaining today had to start somewhere. They learned by doing grunt work that AI now handles.
What's not in dispute: entry-level hiring at AI-exposed firms has fallen roughly 73% by some measures, even as demand for AI-skilled entry-level workers has risen — a contradiction one industry analysis described as companies wanting "entry level" pay with "mid-level" capability. That gap is the actual crisis, regardless of which side of the interest-rate argument turns out to be right.
15 jobs gaining ground
Built by AI directly
- AI/ML engineers represent the fastest-absolute-growth technical hire, with year-over-year AI/ML hiring up 88% in 2025 according to compensation data firm Ravio.
- Data center technicians and operations staff back the more than 600,000 new AI-enabled data center jobs LinkedIn's labor data identifies globally.
- AI Engineers specifically rank as one of LinkedIn's fastest-growing job titles of the past three years.
- Forward-deployed engineers — embedding directly with client teams to implement AI tools — are a category that barely existed before 2023 and now shows up as a named growth role in WEF's 2026 figures.
- Data annotators and labelers, the unglamorous work of training the models, are growing fast enough to be named explicitly alongside AI Engineer roles.
- MLOps specialists overlap with software engineering but focus on deploying and monitoring models already in production rather than building new ones.
- Big data specialists rank among the WEF's top three fastest-growing job categories in percentage terms.
- Cybersecurity analysts and security management specialists are growing as AI both creates new attack surfaces and gets deployed as a defense layer.
Growing because AI can't do them, not because it built them
- Radiologists remain in shortage, with roughly 4,300 unfilled positions and a 130-day average time to fill each one.
- Registered nurses sit inside a profession facing a shortage measured in the millions, where AI handles charting and frees up time rather than replacing the role.
- Skilled electricians and tradespeople are in growing demand as data center construction and the broader infrastructure buildout require hands-on labor AI cannot perform.
- Fintech engineers rank in the WEF's top three fastest-growing roles alongside big data specialists.
- Renewable energy and EV systems engineers are growing on a track largely independent of the AI boom, driven by the parallel green-energy transition.
- Care economy roles — eldercare, childcare, education support — show sustained hiring growth tied to demographics that AI doesn't touch.
- Compliance and risk specialists in AI governance are an emerging category as regulation catches up to deployment, a job that effectively didn't exist five years ago.
What makes the second group interesting isn't that AI spared them. It's that AI made the shortage worse in ways that increased their leverage. Mayo Clinic's radiology department now runs more than 250 AI models and has grown its radiologist headcount 55% since Hinton's prediction — not in spite of the AI tools, but partly because better diagnostic throughput increased demand for the judgment calls only a physician can make on top of it.
The Hinton problem
It's worth sitting with why the most technically credentialed person in the room got this specific call wrong, because the error pattern repeats across nearly every confident jobs prediction since.
Hinton conflated a task with a job. Reading a scan is a task. Radiology is consulting with referring physicians, choosing which imaging protocol fits an ambiguous case, performing image-guided procedures, and explaining findings to a frightened patient sitting in front of you. AI got dramatically better at the first thing. The job is mostly the other things.
Here's the failure mode advocates on both sides of the jobs debate keep avoiding: treating "AI can do X" as equivalent to "AI will eliminate the role that includes X." Nvidia's CEO made this point directly, arguing radiology doomers conflate scan-reading with the entire job — and then, less convincingly, implying the same logic protects every profession equally. It doesn't. Telemarketing really is mostly the automatable task. Radiology really isn't. The honest version of this analysis requires checking, occupation by occupation, what fraction of the job is the part AI is good at — and that's exactly the analysis most "X jobs AI will kill" lists skip, because it's slower than ranking by vibes.
It will not feel slower to the person whose job actually was mostly the automatable part.
Who actually needs to act on this
You're 23, you graduated with a computer science degree eighteen months ago, and you're on your fourth round of "we went with someone with more experience" — for a job described as entry-level. The honest read on your situation, per both sides of the Stanford dispute, is that you're caught in overlapping headwinds: a hiring slowdown that may ease with interest rates, layered under a structural shift that probably won't. Both things can be true, which means the fix is the same either way — build something AI can't yet replicate cheaply, which usually means judgment under ambiguity, not another credential that proves you can do the automatable part well.
You're mid-career in a role from the first list above — bookkeeping, paralegal work, content production. The PwC data on "professionalized" versus "democratized" roles matters here: jobs reshaped to require more human judgment are growing twice as fast and paying 42% more in wage growth than jobs AI has simply made easier for anyone to do. The move isn't fleeing the field. It's moving toward the judgment-heavy end of it before the easy end gets fully absorbed.
You're a hiring manager at a company two years into AI adoption, watching headcount-per-output fall and wondering whether to formalize that into layoffs. The PwC and WEF data both point the same direction: the firms posting the biggest wage and headcount gains are using AI to expand what the business does, not just to do the same thing with fewer people. Cost-cutting alone correlates with the weaker outcomes in this dataset.
The thing rarely said out loud in these conversations: most companies cutting headcount and citing "AI efficiency" are doing something simpler — using a defensible-sounding reason for a decision a board already wanted to make. Not every AI-attributed layoff is actually about AI capability.
Verdict
The net-jobs number is real, and it's also the wrong number to plan your life around. What matters is whether your specific role is mostly one repeatable task or mostly judgment applied to ambiguous situations — and almost nobody's job is purely one or the other. The actual move is auditing your own task list with the same honesty Hinton failed to apply to radiology: which parts of what you do are the automatable slice, and which parts are the consulting-with-the-referring-physician slice. Build toward the second. The first one is shrinking regardless of what title sits on top of it.
Nobody knows yet whether the Stanford Review's interest-rate explanation or the Digital Economy Lab's structural-displacement explanation is the real story behind the entry-level collapse. The graduates living through it don't have the luxury of waiting for that argument to resolve.
FAQ
Will AI actually create more jobs than it destroys?
By the net global figures the World Economic Forum publishes, yes — 170 million created against 92 million displaced through 2030. That net number says nothing about whether the new jobs match the skills, location, or pay grade of the people losing the old ones, which is the part that actually determines whether any individual person is fine.
Is my job actually at risk from AI or is this overhyped?
Check what fraction of your role is one repeatable, structured task versus judgment applied to ambiguous, relationship-dependent situations. Goldman Sachs estimates only about 2.5% of US employment is at meaningful risk if today's AI use cases were rolled out everywhere — far below the doom headlines, but not zero, and concentrated heavily in roles with the first profile.
Why did Stanford's AI jobs study get disputed?
The Digital Economy Lab found a 16% relative drop in employment for 22- to 25-year-olds in AI-exposed jobs. The Stanford Review argued the same data window overlaps with interest rate hikes that hit tech hiring before, citing a near-identical pattern after the dot-com crash that reversed once rates eased. Both groups are using the same numbers to reach opposite conclusions about causation.
What jobs are completely safe from AI?
None are completely safe, but radiologists, nurses, electricians, and care-economy roles have all seen demand and pay rise despite — or because of — AI adoption in their fields. The common thread is tasks that mix physical presence, real-time judgment, or trust built over time, none of which current AI does well.
Should I avoid majoring in computer science because of AI?
Enrollment has already dropped at several University of California campuses over this exact fear. The counterargument worth weighing: entry-level coding tasks are what's shrinking, not software engineering broadly, and the WEF still projects an 82% increase in machine learning roles. The risk is graduating with only the automatable skill set, not the degree itself.
Is the AI jobs crisis the same as past automation waves?
Partially. Accountants were predicted to vanish when spreadsheet software arrived in the 1990s; instead, routine number-crunching disappeared and the role shifted toward advisory work. The open question economists disagree on is whether generative AI's task range is broad enough to break that historical pattern rather than repeat it.
How fast is the AI jobs skills gap actually widening?
PwC's 2026 analysis of more than a billion job postings found the skills required for AI-exposed roles are changing more than twice as fast as for sheltered roles — a gap that grew 75% in just one year. That acceleration, not the net jobs number, is the figure to actually watch.