The most-quoted prediction about AI and jobs turned out to be wrong. The data is teaching us a different story, if we look.
In 2016, the computer scientist Geoffrey Hinton (sometimes called "the godfather of AI") stood on a stage and said something that, in hindsight, has aged about as well as any prediction in modern technology:
"I think if you work as a radiologist, you are like the coyote that's already over the edge of the cliff but hasn't yet looked down. People should stop training radiologists now. It's just completely obvious that within five years, deep learning is going to do better."
Ten years later, the U.S. Bureau of Labor Statistics projects radiology employment to grow 9% through 2034, three times the rate of the average occupation. The American College of Radiology counted 4,333 active radiologist job listings in March 2026. Average radiologist salary, per Fortune, reached $571,000, up 9% year over year. Open positions take an average of 130 days to fill.
You can read those numbers two ways.
You can read them as a single anecdote: one prediction missed; the man who said it is still brilliant; AI is still going to come for everyone else. Or you can read them as a pattern that repeats: predictably, mechanically, across every wave of automation in the last 200 years.
This piece is about why the pattern keeps repeating. And why, for most people reading this, the perspective they choose to take on it will matter more to their career than the underlying technology will.
The ATM story
If you've heard one historical parallel for AI and jobs, it's probably the ATM. Bank tellers were supposed to disappear when machines started counting money. They didn't. The Boston University economist James Bessen documented what happened instead: between 1988 and 2004, the number of tellers per branch dropped by more than a third. But the number of urban bank branches grew 40%. Net effect: the number of bank tellers in America doubled during the years ATMs were rolling out nationwide.
The ATM didn't kill the teller. It made each teller cheaper to deploy. So banks deployed more of them, in more places, doing different work. Routine cash handling went to the machine; relationship banking, financial product sales, and complex customer service became the new core of the job.
When the teller population finally did decline, in the 2010s, it wasn't ATMs that caused it. It was mobile banking. A different technology, in a different decade, doing a different thing to the underlying economics. The thing everyone feared in 1980 was the wrong thing.
That same shape, the same wrong-thing-feared shape, is playing out right now with AI and radiologists.
What actually happened to radiology
When you compress what AI can do for an imaging study (flagging lesions, measuring densities, prioritizing the queue, drafting first-pass reads), you don't get a world with fewer scans. You get a world with more scans, because each scan got cheaper to perform and interpret.
The U.S. now reads about a billion diagnostic imaging studies per year, growing at roughly 3-4% annually. The drivers are mostly demographic and structural: an aging population, expanded cancer screening protocols, growing utilization of CT and MRI in emergency medicine, and the explosion of image-guided interventional procedures. AI made each scan cheaper, which expanded who got scanned, which created more interpretation work, not less.
This is sometimes called the Jevons Paradox in economics: when efficiency goes up, the resource often gets used more, not less. Cheaper steam engines used more coal. Cheaper compute uses more compute. Cheaper imaging means more imaging.
CNN called radiology "the ultimate case study for why AI won't replace human workers." Reed Smith framed it as the field where AI augments rather than replaces. The data is the data.
What's happening to software engineers
The other field everyone says is being eaten by AI: software engineering. The narrative is loud: 52,000 tech layoffs in Q1 2026, nearly half attributed to AI. The narrative is also incomplete.
In the same quarter:
- Software engineer job listings jumped 30% in 2026, with more than 67,000 openings tracked across major tech companies, the highest demand in over three years.
- Indeed software engineer postings are up 11% annually, a faster clip than postings overall, per Citadel Securities' tracking.
- "AI engineer" was LinkedIn's #1 fastest-growing job title in the U.S. in 2026. Year-over-year postings: up 143%.
- Prompt engineering demand grew 135.8% in one year.
- Workers with AI skills are earning a 56% wage premium over peers.
- The BLS projects computer and IT occupations to grow faster than average through 2034, with about 317,700 openings projected per year.
What's happening is a redistribution, not a collapse. The roles AI eliminates and the roles AI creates aren't the same roles. A backend engineer maintaining a CRUD app may have a harder time than they did in 2022. An engineer who can build, deploy, and govern an AI agent is in the most competitive market of the last decade.
Why perspective matters more than the data
Here is the uncomfortable part: every statistic in this article is publicly available. It's been available for a year. The narrative we get in mass media (AI is coming for your job; entry-level work is dead; the great displacement is upon us) isn't being suppressed by the data. It's coexisting with the data.
The fear narrative is louder than the abundance narrative because loss is psychologically more legible than expansion. A laid-off engineer at a major tech company is a story; the 67,000 listings sitting open across other companies is a statistic. The mind weighs them differently. Headlines weigh them differently. Most of what shapes our collective sense of the moment comes from the difference in how those two facts feel, not how they measure.
This is the perspective trap. Not that we have bad data. That we don't notice we're choosing which data to feel.
If you spend a year telling yourself "AI is destroying my profession," you will see every layoff confirming it and miss every listing contradicting it. If you spend the same year asking "where is this making work cheaper, and what gets unlocked at the new price?", you'll find a different career.
Both observers are looking at the same world. They're just looking through different frames.
What this means practically
If you take one thing from this piece:
Look at what's growing, not just what's shrinking. Every productivity wave in the last 200 years has had the same shape: a small number of jobs disappear loudly, a much larger number of new jobs appear quietly, and the people who position themselves on the growth side of the wave do better than the people who position themselves against it.
The 2016 prediction got it wrong because it treated AI as a substitute for human radiologists. AI turned out to be a complement: a force that lowered the cost of each scan, which expanded total demand for scans, which expanded total demand for radiologists.
The same thing is happening, on a longer arc, in software. It's starting to happen in legal services, accounting, marketing, design, and customer support. It will happen in dozens more fields before the decade ends.
What you see depends on what you're looking for. The data is there. The question is which story you're choosing to tell yourself about it.
Sources
- BLS Employment Projections, 2024-34. U.S. Bureau of Labor Statistics
- Fortune: "A decade after the 'Godfather of AI' said radiologists were obsolete, their salaries are up to $571K and demand is growing fast" (May 2026)
- CNN Business: "Worried about AI replacing your job? This job has become the ultimate case study for why it won't" (Feb 2026)
- Reed Smith: "Why is AI not replacing the demand for radiologists' services?"
- CNN Business: "The demise of software engineering jobs has been greatly exaggerated" (April 2026)
- Metaintro: "Software Engineer Job Listings Are Up 30% in 2026"
- James Bessen: "Toil and Technology: ATMs and bank tellers," IMF Finance & Development
- American Enterprise Institute: "What the Story of ATMs and Bank Tellers Reveals About the 'Rise of the Robots' and Jobs"
- BLS: "AI impacts in BLS employment projections"
- HeroHunt: "Fastest Growing AI Roles in 2026: Data and Rankings"