In my last post I talked about how building with AI feels amazing and terrifying at the same time. This one is about the terrifying part. Specifically, what do we actually do about it?
Because I keep hearing the same reassuring answer and it doesn’t hold up.
“They’ll just move to trades”
This is the comfortable narrative: sure, AI will automate white-collar work, but there’s a skilled trades shortage! People will become electricians, plumbers, welders. Problem solved.
I think this is dead wrong, and the math is easy enough that even I can do it.
There are roughly 70 million workers in management, professional, and related occupations in the US. The entire skilled trades workforce (construction, manufacturing maintenance, utilities) is about 12 million. That’s a 6:1 ratio.
Goldman Sachs estimates AI could displace 9 to 10 million US workers. Vinod Khosla thinks AI will be capable of doing 80% of all jobs by 2030, affecting $15 trillion of US GDP in labor. Even if you take the conservative end, say 5 million displaced white-collar workers, that’s nearly half the entire existing trades workforce trying to squeeze into a sector that currently has about 1 to 1.5 million unfilled positions.
And that’s just in the US.
And those positions aren’t “show up Monday” jobs. Becoming an electrician takes 4 to 5 years of apprenticeship and 8,000 to 10,000 hours of combined training. Plumbing is similar. You can’t retrain a project manager into a journeyman electrician over a long weekend. The pipeline to even absorb those workers doesn’t exist, and building it takes years. Years we may not have if Suleyman’s “12 to 18 months for white-collar task automation” forecast is even partly right.
Here’s what I think actually happens: 90 of the previous 100 software engineers decide to go be electricians. What you get isn’t a revitalized trades sector. What you get is the collapse of trades wages. Massive labor oversupply craters bargaining power. Union density in construction has already fallen from 90% in the 1930s to a record low of 10.7% in 2023. Flood the sector with desperate white-collar refugees and whatever collective power remains evaporates. Everyone competes on price. Nobody wins.
We’ve seen this movie before
This isn’t speculation. We have the data from manufacturing.
When 5 million manufacturing jobs disappeared between 1998 and 2019, displaced workers didn’t smoothly transition into new careers. Only 42% were reemployed at all, the lowest reemployment rate of any sector. Of those who did find work, 65% earned less than before. The average annual pay cut was over $10,000. Many just exited the labor force entirely.
And that was a slower disruption than what AI is doing. Manufacturing decline played out over two decades. The AI displacement forecasts we’re talking about are 1 to 5 years.
The cascade nobody’s talking about
But it gets worse, because it’s not just a labor problem. It’s a demand problem.
If millions of people lose their income or take massive pay cuts, who buys the things AI is producing? Who buys houses? Who goes to restaurants? Who buys the products these newly efficient companies are making?
A viral essay from Citrini Research in February 2026 called this “Ghost GDP,” economic output that inflates national accounts but never circulates through the real economy because machines don’t spend money on goods and services. The argument: companies replace workers with AI to protect margins, displaced workers lose purchasing power, consumer spending drops, companies need more cost-cutting, more AI adoption, more displacement. A deflationary spiral.
Citadel pushed back, arguing that productivity gains lower prices and increase real purchasing power. And maybe that’s true in a gradual transition. But if 10 to 20% unemployment hits within a few years, which is what Amodei, the CEO of the company whose AI I use every day, is publicly warning about, “lower prices” isn’t going to matter much to people who can’t make rent.
Non-housing household debt is already at record levels. Roughly $5 trillion in 2025, largely credit cards. Real wages for the middle quintile grew only 2.6% from 2019 to 2024, the weakest of any income group. The middle class is already stretched. AI displacement doesn’t hit a resilient economy. It hits one that’s already fragile.
The people who built this are worried too
Look, I’m not some Luddite who thinks we should ban AI. I use it every day. I love building with it. But the people who created these systems are saying the quiet part out loud now.
Dario Amodei warned in May 2025 that AI could wipe out half of entry-level white-collar jobs and push unemployment to 10 to 20% within one to five years. He said AI companies and governments need to stop “sugar-coating” the risk.
Mustafa Suleyman said most computer-based white-collar tasks could be automated within 12 to 18 months. Back in 2023 he was already saying AI would create “a serious number of losers” and governments should think about UBI.
Geoffrey Hinton told the UK government in 2024 that UBI was a good idea. By 2025 he’d walked it back. Not because the displacement risk was smaller than he thought, but because he realized cash alone doesn’t solve the loss of dignity and purpose that work provides. Money helps. Money isn’t enough.
Then you’ve got Jensen Huang saying AI infrastructure will create plenty of jobs. Sure, but the CSIS found that apprenticeships need to expand 50% by 2030 just to meet AI infrastructure demand, and one-third of the current skilled labor force is over 50 and retiring. Who trains the trainers?
We need UBI. And we need guardrails.
So yeah, I think we need universal basic income. Not as the whole answer, but as a floor. Because the alternative, tens of millions of people competing for a shrinking pool of jobs while a handful of companies capture most of the economic value, is a recipe for the kind of social instability that doesn’t end well for anyone, including the companies.
But UBI alone isn’t enough. Hinton is right about that. People need purpose. They need agency. A subsistence check and a pat on the head is not a society.
We also need guardrails on AI itself. Not “ban it,” that ship has sailed and I wouldn’t want it to anyway. But some kind of framework for how fast displacement can happen, how gains are distributed, what obligations companies have when they automate away roles. Sam Altman talks about “universal extreme wealth” and ownership-style sharing of AI gains. I’d settle for “universal not-falling-off-a-cliff.”
AI is the most powerful tool I’ve ever used. The ability to dream up something and have a working prototype in an hour is genuinely magical. I’m having the time of my life. And I’m watching a slow-motion economic crisis that we’re choosing not to prepare for because the stock market looks fine and the people getting displaced aren’t loud enough yet.
We need UBI. We need guardrails. We need to stop pretending “learn to code,” or now “learn to weld,” is a policy. And we need to do it before the cascade hits, not after.