AI & Inequality: New Year’s Eve Edition 🎆

The cognitive elite are about to learn what factory workers discovered in the 1980s: your job wasn’t protected by your skills. It was protected by the cost of replacing you. And that cost just collapsed.

Welcome to 2026. White-collar work faces its reckoning. AI doesn’t trim the margins. It redefines the core.

The IMF’s Uncomfortable Revelation

Recent IMF research reveals something unusual about AI compared to previous automation waves: it may reduce wage inequality by displacing higher-income workers more than low-income ones. This isn’t because AI is kinder to the working class. It’s because the “non-routine cognitive tasks” that justify six-figure salaries turn out to be remarkably routine when you have a multimodal model with access to the entire corpus of human knowledge.

Think about what we’ve been calling “knowledge work” for the past forty years. Legal research. Financial analysis. Strategic consulting. Management reporting.

These weren’t actually unique cognitive achievements. They were pattern matching at scale, performed by humans because computers couldn’t do it yet.

Until now. Allegedly.

The Prophecy Industrial Complex Strikes Again

Dario Amodei, CEO of Anthropic, estimates that nearly half of all entry-level white-collar jobs in tech, finance, law, and consulting could be replaced or eliminated by AI. It’s a bold claim from someone whose business model depends on corporate clients believing exactly that.

Let’s be clear: AI is impressive. It’s also spectacularly bad at things humans handle effortlessly.

“The models got smarter, benchmarks improved, and the infrastructure scaled impressively. None of it was enough to bridge the gap between ‘this is remarkable’ and ‘this is AGI.’”

ChatGPT still hallucinates references roughly one in ten times. GPT-4 maintains a 28.6% hallucination rate in medical systematic reviews. These aren’t minor glitches when you’re drafting legal briefs or analyzing portfolio risk.

The AI prophets conveniently ignore this when forecasting mass displacement. It’s easier to sell software licenses when clients believe the alternative is obsolescence.

The Class Drama Inside the Knowledge Economy

But here’s where it gets interesting. The displacement—when it happens—isn’t hitting equally across the income spectrum.

Picture a law firm. Not the romanticized version from television, but the actual economic structure. Partners at the top, associates in the middle, paralegals at the bottom. The entire system worked because each tier added value the tier below couldn’t provide.

London-based law firm A&O Shearman has built AI tools that scan 20 years of license agreements and identify which need amending—work that two years ago would have required 20 lawyers in a room. The associates who would have spent billable hours on that work? They’re discovering their “non-routine cognitive labor” was actually quite routine after all.

The same pattern is replicating across investment banking, management consulting, and corporate strategy. Roughly one-third of functions in an average U.S. job can be augmented or automated by generative AI, with that proportion growing sharply in data-heavy sectors like finance, legal services, and consulting.

“What’s fascinating is the inversion. The paralegal filing documents? Still employed. The junior associate doing document review? Endangered. The managing director overseeing relationships and negotiating deals? Probably safe.”

For now.

We’re watching a class drama unfold in real-time. The top 10% of earners—the ones who spent two decades assured their education and professional credentials were insurance against disruption—are discovering that insurance has an expiration date.

Assuming, of course, that AI can actually deliver on its promises. Which remains an open question given the current state of “AI slop” flooding the market.

The Real Line of Defense: Capital, Not Credentials

Here’s where the IMF research gets brutal. While AI could reduce wage inequality through displacement of high-income workers, a few factors counter this effect. If you’re a high earner who can use AI to amplify your productivity, you might survive. If you own the companies deploying AI, you’ll thrive. If you’re just selling your cognitive labor hour by hour? You’re about to become the knowledge economy’s equivalent of a factory worker.

The top 20% wealthiest U.S. households own nearly 93% of all stock, meaning they get the lion’s share of any stock market gains. The stocks of companies tied to artificial intelligence have accounted for roughly 75% of S&P 500 returns since ChatGPT launched in November 2022.

The wealth gap isn’t widening because of wages. It’s widening because of ownership.

“The IMF found that while AI may reduce wage inequality by displacing high-income workers, it is likely to substantially increase wealth inequality as these same workers benefit from higher returns on their capital holdings.”

The partner with equity in the firm survives. The associate paid in salary? Disposable.

This is the class structure emerging in the AI economy:

The Capital Class: Own AI companies, own equity in firms deploying AI, capture returns on capital as productivity soars (theoretically)

The Complementary Class: High-skill workers whose tasks AI enhances rather than replaces—the surgeons, the dealmakers, the creative directors who fix AI’s mistakes

The Displaced Class: Everyone whose “knowledge work” turns out to be just expensive pattern matching (or would be, if AI could actually do it reliably)

The Cognitive Assembly Line (That Keeps Breaking Down)

In 2025, hiring for professional roles in finance, technology, consulting, marketing, and law slowed dramatically or stopped altogether, even as corporate profits remained robust and productivity soared. This isn’t a recession. It’s a restructuring.

Or perhaps it’s corporations discovering they can freeze hiring, blame AI, and pocket the savings while the technology catches up to the hype. Either way, the workers lose. The machines aren’t just getting better at rote tasks. They’re getting better at the tasks we used to think required human judgment.

Except when they confidently hallucinate case law that doesn’t exist. Or misinterpret regulatory guidelines. Or fail to catch the kind of contextual nuance that separates competent analysis from career-ending mistakes.

The irony is exquisite. For decades, the professional class watched manufacturing jobs evaporate and consoled themselves with the belief that their cognitive skills were immune to automation.

“Learn to code,” they told displaced factory workers. “Get a college degree. Do knowledge work.”

Now they’re discovering that knowledge work is just another assembly line. And the assembly line is being automated. Imperfectly. Expensively. But automated nonetheless.

What’s Next?

When firms can choose how much AI to adopt, the wealth inequality effect is particularly pronounced, as the potential cost savings from automating high-wage tasks drive significantly higher adoption rates.

Every CFO in America is doing the math right now: replace three $150,000 associates with one AI license and one partner. The productivity stays the same. The cost drops by half.

In theory.

In practice, someone still needs to check whether the AI just cited a case from a parallel universe or recommended a tax strategy that violates three different securities laws. But that someone costs less than three associates, so the math still works.

Companies will shift from humans to machines the moment these technologies can operate at a human efficacy level. This could eliminate tens of millions of jobs in a short period. Or it could take decades. Or it could plateau at 70% efficacy, requiring permanent human oversight. The prophets don’t mention that scenario because it doesn’t sell consulting contracts.

The top 10% will split into two groups: those who own capital and those who don’t.

The former will capture unprecedented wealth as AI-driven productivity compounds their returns. The latter will discover that their professional credentials, their graduate degrees, their years of experience don’t protect them from the same forces that hollowed out Rust Belt manufacturing towns.

The labor class loses. The capital class wins.

Whether the machines can actually deliver remains an open question. But the ownership structure? That’s already locked in.

The factory workers saw it coming but couldn’t stop it. Now it’s the professionals’ turn. The only question is whether they’ll learn the lesson faster: in the age of AI, your job title doesn’t matter. Your ownership stake does.

Happy New Year. May your 2026 be filled with common sense—the one thing AI still can’t replicate.

For more insights about what AI can or cannot do, check out my book “Artificial Stupelligence: The Hilarious Truth About AI”.