The Demise of the Prestige of White-Collar work (RIP)
White-collar work is the new blue-collar work. And blue-collar work might survive AI.
‘We’ decided the most automatable work in the history of capitalism was the most prestigious. Somehow, that was always going to end badly.
We built an entire class mythology on the wrong assumption: “thinking” was harder to replace than “doing”. It turns out thinking, especially when it happens on a computer in a structured environment, is exactly what machines are built for. ‘Doing’, when it happens on a roof in the rain or under a leaking crawlspace with a flashlight, is not.
The knowledge class is about to find out what the working class has always known: no credential protects you from a cheaper version of yourself.
We want to be precise about what this is and isn’t, because the story is more unsettling than “AI takes your job.” And we’ll tell you exactly what to do about it. But first, you need to understand how we got here.
The Prestige Trap
For a hundred years, white-collar workers lived inside The Prestige Trap, the self-reinforcing belief that cognitive work was categorically different from physical work, and therefore categorically safer. The trap had two jaws: a credential system that made the belief feel earned, and an economy that rewarded it long enough to seem permanent. Both jaws of the trap are now open.
Upton Sinclair, the man who coined the term “white collar” in 1919, saw it first. The white-collar worker, he wrote, was “often the worst exploited of proletarians,” performing middle-class identity on a working-class wage. In 1951, sociologist C. Wright Mills made it the subject of a whole book: a new class of workers with no independent power base (no property, no union, no political identity) whose entire sense of self-worth rested on a status they did not actually control. The credential was a costume. The security was borrowed. Both men saw it coming. Neither got the timing right.
Thanks to AI, the time is now. Just think of the fright the last couple of years has given even the most prestigious of workers - finance and tech workers - and how they have felt super exposed.
Outsourcing was the Proof of Concept
The Prestige Trap held as long as cognitive work was hard to move. The moment it became moveable, leaks appeared.
The rehearsal was outsourcing. Beginning in the 1990s, white-collar work began its first migration: call centers to Manila, software to Bangalore, legal research to Mumbai. What we failed to apprehend was that outsourcing wasn’t a crisis. It was a proof of concept.
The moment knowledge work became transmissible over a wire, it became subject to commodity logic: find the cheapest source, extract the most output, then repeat, repeat, repeat. The offshoring wave climbed the value chain steadily. Junior legal hours at U.S. law firms dropped 60% in the five years leading up to 2014. The jobs that were supposed to be safe because they required intelligence turned out to be unsafe for exactly that reason: intelligence, in structured environments, travels well and has no face.
AI is offshoring at the speed of light, with no visa required.
History Says More Work Will Emerge. But that Misses the Point.
Every major automation wave in history has ended with more work, not less. The spreadsheet didn’t eliminate accountants. It created millions of them. The internet didn’t kill publishing; instead, it detonated the number of independent publishers. Thirty million professional developers were a ceiling until AI coding tools arrived; now, 150 million people may write code who never called themselves developers. Vibe coding is not the death of engineering. It is the democratization of it.
AI will likely follow this pattern. History tells us work expands. People think it might be different this time. Time will tell.
What we do know is that The Great Inversion is not about the quantity of white-collar work. It is about its standing. Its wage premium. Its prestige. The compact that said: get the degree, do the cognitive labor, earn a certain kind of life.
That compact is what AI is breaking. Not the work itself, but instead the deal attached to it. The factory didn’t disappear in the 1970s, but the dignity attached to factory work did. That is the closer analogy, and the one that should make every knowledge worker a bit twitchy with dread.
Geoffrey Hinton won the Nobel Prize for inventing the technology that is about to remake white-collar work. In 2025, a reporter asked him which jobs were safe. He said: plumbers. Not AI researchers. Not lawyers. Not analysts. Plumbers.
The man most responsible for building the AI reshaping the knowledge economy, the man they call the Godfather of AI, looked at what he created and pointed to the guy with the wrench. That tells you everything about where this is going, and it points directly to why.
Blue-collar work has a bigger moat. For now.
Blue-collar work has always had an accidental advantage nobody bothered to pinpoint precisely. We call it The Embodiment Moat: the structural protection that physical, situated, unpredictable work enjoys precisely because it cannot be transmitted over a wire. You cannot email a crawlspace. You cannot Slack a roof in a rainstorm. The moat wasn’t built intentionally. But it turns out to be the only one that matters in the age of AI. In fact, due to a shortage of workers, many tradespeople are seeing annual salary increases of around 3.8%. Construction pay, in particular, has seen substantial gains, with average hourly earnings in the US reaching nearly $40 by August 2025. Specialized roles, such as electricians and HVAC technicians, are now frequently seeing six-figure salaries.
A large language model doesn’t need a body to be effective. It can do in seconds what a junior analyst does in a week: reading documents, synthesizing data, drafting arguments, and running scenarios. Goldman Sachs found that AI systems can now match or outperform up to 47% of industry professionals on economically valuable tasks. McKinsey estimates that today’s existing technology could automate 57% of current U.S. work hours, not in theory, not in ten years, but now. We generally take these self-interested statements with a grain of salt. However, entry-level hiring in AI-exposed jobs has already dropped 13%. Internship offers from Fortune 500 companies fell 22% between 2022 and 2024.
And in case you’re wondering, OpenAI is openly tracking how much of GDP-involved work can be done by the GPT series of models. Go check our GDPVal.
Meanwhile, 500,000 net new trade jobs need to be filled to meet U.S. power demand by 2030. Construction jobs tied to the data center build-out (the physical infrastructure of the AI economy) have increased by 216,000 since 2022. The workers building the systems that will displace knowledge workers are protected by those systems’ own limitations. The Embodiment Moat holds.
There’s a New ‘Collar’ class emerging
History’s optimism about work volume is probably right. But the new prestigious work won’t look like the old prestigious work.
A genuinely new category of worker is forming. We call them The Conductor Class: people whose primary skill is not doing cognitive work directly, but directing the artificial intelligences that do. Not building AI like the labs. Not operating it mechanically. Conducting it: shaping intent, setting constraints, evaluating output, pointing multiple AI systems toward outcomes that require human judgment to define and human accountability to own.
This may be the central professional role of the next thirty years. Every organization will need people who can tell AI what to do, evaluate whether it did it well, and fix the system when it didn’t. That requires domain expertise, taste, and judgment, skills that are hard to automate because they are, by definition, about directing the automator.
The Conductor Class is NOT the old white-collar class. Its power comes not from knowing a domain exhaustively, but from knowing it well enough to direct an intelligence that can know it exhaustively.
In fact, having specific subject-matter knowledge in a single domain makes you brittle and vulnerable. Unless you are literally creating meaningful knowledge on a global level, you will be protected only until the economics make your expertise worth automating. The bar shifts from depth of knowledge to breadth and quality of judgment. The judgment you developed by building specific domain knowledge is what’s important, not the knowledge or the use of the knowledge itself. The value shifts from execution to orchestration and from doing to conducting.
That is a real and valuable job. But it will not command the same premium by default, or simply because you have a degree and showed up. The premium will go to those who combine training with the development of genuine orchestration skills. The credential doesn’t transfer automatically.
What To Do About It
Here are four specific moves, not career advice per se, but decisions you can make in the next 100 days.
Run your exposure audit. Map your actual working week. How much of what you do is transmissible over a wire: document production, analysis, synthesis, research, drafting? That ratio is your AI exposure score. If it is above 60%, you have a structural problem worth solving now, not at your next annual review. Be honest. Most knowledge workers, when they map it clearly, are sitting at 70% or above.
Move one responsibility from execution to evaluation. The work that survives AI is judgment, not production. If you are still the person generating the first draft, running the first analysis, or building the first model, that is your exposure. Pick one of those and flip it: use an agent to generate, and make yourself the one who defines the brief, evaluates the output, and owns the decision. Do that once, deliberately, and learn what it actually feels like to conduct rather than perform. Then do it again.
Develop Conductor Class skills: not prompting, orchestration. Prompting is the floor. The ceiling is workflow design, output evaluation, and the domain expertise to know when AI is wrong in ways a non-expert wouldn’t catch. That last part is the moat. A lawyer who can evaluate AI legal research better than a non-lawyer is valuable. A lawyer who still manually does legal research is expensive. Find the equivalent in your domain and build toward it.
Identify your baton. Every Conductor has a baton: the specific domain expertise that makes their AI supervision worth paying for. What do you know that makes you the right person to direct an AI in your field? Name it explicitly. It is probably not your title or your degree. It is the hard-won judgment about what good looks like in your specific domain, the thing you built from years of doing the work that AI is now doing faster. That judgment is not obsolete. It is the most valuable thing you own right now. Sharpen it.
White-collar work is not disappearing. The mythology about it being better than blue-collar work is. The question is not whether you will have a job. It is whether you will conduct the intelligence or be replaced by the person who does.
Part II, for organizational leaders managing teams through this transition, will be published next week.
🔍 Want to dive deeper?
Check out our book BUILDING ROCKETSHIPS 🚀 and continue this and other conversations in our 💬 ProductMind Slack community and our LinkedIn community.
P.S. AI models just leaked.
We break down what it means and why it matters more than people think.
🎥 YouTube → Click Here
🎵 Spotify → Click Here
🎙️ Apple Podcasts →Click Here
📢 We’re also excited to share that Ted Yang has a new book coming out:
📕 Ageless Peak Performance: The Playbook for AI-Powered Excellence.
In it, Ted explores how AI can amplify human capability and help us reach new levels of performance.
Pre-order now HERE 🎉