Nick Manella: The Doom Is Mostly Vibes

Spend a week reading tech press right now, says Nick Manella of New Hope, PA, and you’ll come away convinced the labor market is on fire. AI is taking your job. Software engineering is dead. White-collar work is finished. Soon we’ll all be living in pods while seven guys in Atherton split the rest of the economy between them.

Nicholas Manella wants to take this seriously because he knows that parts of it are real. But most of it isn’t. And the gap between what the headlines say and what the actual labor data shows has gotten wide enough that somebody owes somebody an apology, and he doesn’t think it’s the economists.

Below, Nick Manella walks us through the truth of the matter.

The number nobody quotes

Here’s the single most important statistic in the AI jobs debate, says Manella. People almost never see it in a headline:

Net increase of 78 million jobs globally by 2030.

That’s from the World Economic Forum’s Future of Jobs Report 2025, which surveyed more than a thousand of the world’s largest employers across 22 industries. Their projection: 92 million jobs displaced, 170 million new ones created, net gain of 78 million.

This isn’t some optimist blogger’s hot take. It’s the consensus institutional forecast – the same body that runs Davos. And as Nicholas Manella knows, it lines up with everyone else doing serious modeling. Goldman Sachs, WEF, IMF, McKinsey, the BLS – every major institutional projection shows net positive job creation through the AI transition.

Meanwhile? 63% of American workers believe AI will decrease overall job availability.

There’s a hundred-million-job gap between what regular people are afraid of and what the actual forecasters, including the ones building and selling AI, are projecting. Somebody’s wrong, says Manella. And so far, the data isn’t backing the doom side.

Okay, but what’s actually happening right now?

Let me steelman the panic for a second, says Manella, because real things are happening to real people.

Workers aged 22–25 in the most AI-exposed roles have seen a 6% drop in employment from late 2022 to September 2025. Software developers in that same age bracket saw an almost 20% decline from their late-2022 peak. U.S. tech layoffs hit 170,630 in 2025. If you’re a recent CS grad in 2026, the job market feels nothing like what your professors described in 2021. That’s real, says Hope, PA’s Nick Manella.

But look at what’s actually going on. Says Manella. AI is suppressing hiring more than it’s destroying existing jobs. Goldman Sachs reads it as employers integrating AI to avoid adding headcount rather than firing people they already have.

That distinction matters a lot. The pain is concentrated in one cohort – young people trying to break in – not spread across the whole economy. Among 22–25-year-olds in AI-exposed roles, employment fell 16% from late 2022 to mid-2025, while experienced workers in the same fields stayed largely stable.

That’s a serious problem. It’s also a fundamentally different problem than “AI is taking everyone’s job.” The actual aggregate U.S. number from AI-driven cuts? Around 55,000 jobs through 2025. In a labor market of 160 million workers, that’s a rounding error.

And historically, each one-point productivity gain from technology raises unemployment by about 0.3 points in the short run, and most of those losses fade within two years. Which is roughly what we’re watching unfold in real time.

The bank teller story

There’s a thing economists have been telling each other for forty years that Manella thinks gets at the real shape of what’s happening.

When ATMs came online, the prediction was obvious. Machines that hand out cash will eliminate the people who hand out cash. Makes sense.

Here’s what actually happened. ATMs cut the number of tellers needed per branch from 20 to 13 between 1988 and 2004. So banks opened more branches. Urban bank branches went up 43 percent. Fewer tellers per branch, more branches, total teller employment didn’t budge. It actually grew from roughly 300,000 in 1970 to over 600,000 by the early 2000s, even as ATMs blew past 400,000 nationwide.

This is Jevon’s paradox. Make something cheaper, and people use more of it, not less. Cheaper tellering meant more branches meant more tellers.

As Nick knows, here’s the part nobody mentions: the story didn’t have a happy ending forever. By 2023, U.S. bank branches had fallen to under 78,000, down about 22% from the peak. Tellers got hit. But the thing that killed them wasn’t the ATM. It was the iPhone, which killed the reason anyone walked into a branch.

Manella thinks this is the most useful frame for thinking about AI and any specific job. Task automation grows adjacent roles. Purpose automation kills them. “Can AI do the tasks in this job?” and “Does AI eliminate the reason this job exists?” are two completely different questions, and people who can’t tell them apart are the ones making the worst predictions.

What “57% automatable” actually means

You’ve probably seen this stat or something close to it. Half of all work could be done by AI. Below, Manella shows what the actual research says, because the way it gets reported is almost dishonest.

McKinsey’s number is that today’s technology could, in theory, automate about 57% of current U.S. work hours. That’s not 57% of jobs disappearing. It’s 57% of hours across the entire working population that involve tasks an AI system could, in principle, handle. Deployment is the limiting factor, not capability.

To translate, the AI can technically do the thing in a lab. Whether anyone deploys it, integrates it, gets it to work with the rest of the business, and actually replaces a human task is a different problem entirely.

McKinsey also finds that 60% of jobs have at least 30% of their tasks automatable today, says Manella. But full job displacement is much rarer than task displacement.

Jobs are bundles of tasks. When some tasks in a job get automated, the job usually doesn’t vanish, it shifts. The leftover tasks get more valuable. New tasks show up. Bank tellers in 2026 do more sales and customer service and less cash counting. As Manella knows, radiologists with AI assistance read more scans per hour, not zero scans because they got replaced.

If AI were truly substituting for human labor the way the panic story claims, we’d see wages collapsing in exposed fields, says Nicholas Manella. The Dallas Fed looked at this through early 2026 and found wages in AI-exposed jobs weren’t uniformly declining, suggesting that for most workers, AI is augmenting their output rather than replacing it.

That’s what complementary technology looks like in the data. Hiring slowdown at the entry level, stable wages for experienced workers, productivity going up. It’s not the shape of an apocalypse. It’s the shape of an adjustment.

What’s actually worth worrying about

Manella is not trying to sell pure optimism. The mirror image of doomerism is its own kind of unserious. Three real things to take seriously:

The entry-level problem is bad and probably getting worse. A generation of workers who can’t get a foothold is a slow-motion disaster. Harder to see than mass layoffs, but the long-tail damage to careers and earnings is real. The downstream consequences for workers who can’t find entry-level jobs to start building career capital are potentially as severe as direct displacement, just slower and harder to measure. This needs policy attention now, not in 2030.

The speed thing is genuinely new. The ATM transition played out over forty years. The AI coding transition took about three. When change comes that fast, the economy doesn’t have room for the usual adjustment mechanisms: companies can’t reorganize, new markets can’t emerge, workers can’t retrain. The historical analogies that say “it’ll all work out” assumed decades, not quarters.

Inequality is the real story, not unemployment. Even in the optimistic WEF scenario, 39% of existing skill sets will become outdated between 2025 and 2030, and 63% of employers say skills gaps are their biggest barrier to transformation. The people who own AI infrastructure and the people who can use it well will capture almost all the gains. Everyone else gets squeezed. That’s a fixable problem… but only if we admit it’s the actual problem, instead of staring at the wrong threat.

These are adjustment problems. They’re not obsolescence problems. And the policy responses look completely different. Adjustment problems want targeted retraining money, portable benefits, wage insurance, and antitrust enforcement on AI infrastructure. Obsolescence problems want UBI and panic. Pick the right diagnosis, and the prescription changes.

Why the doom keeps winning the news cycle

There’s a reason the apocalyptic story keeps beating the boring-true story, says Manella, and it has nothing to do with the data.

Run through the incentives. If you’re an AI company, “this is so powerful it will replace human labor” is a fundraising deck. If you’re a politician, “AI is coming for your job” is a fundraising email. If you’re a journalist, “humanity faces extinction-level workforce disruption” gets the clicks. If you’re a doomer pundit, you get a permanent speaking circuit. Everyone in the prediction business has incentives pointing in one direction. None of those incentives is about being right.

The boring truth, as Manella sees it: we’re in a real but manageable adjustment; the labor market is doing roughly what it always does during a technology shock; and we should target policy at the actually affected groups instead of LARPing about civilizational collapse.

This might not move many newspapers, but it’s what the data says.

So what?

Manella keeps this short because he knows the implications fall out naturally:

Stop writing policy around science fiction. Write it around what we can actually measure: collapsing entry-level hiring in exposed fields, geographic concentration of AI layoffs, skill gaps closing too slowly, and returns flowing to capital instead of workers. These are real, specific, and addressable.

Be skeptical of anyone-boomer, doomer, anyone-who tells you they know exactly what AI does to the labor market in 2035. They don’t. Manella doesn’t. Nobody does.

But the data we have in front of us right now, in 2026, looks a lot more like 1995 than 2050. And in 1995, the people who panicked about the internet destroying jobs were extremely, embarrassingly, and expensively wrong.

Sources: WEF Future of Jobs Report 2025; Goldman Sachs Research; McKinsey State of AI 2025; Federal Reserve Bank of Dallas; James Bessen, IMF (2015); Yale Budget Lab; Bureau of Labor Statistics.