AI is already hollowing out the white-collar economy

On a Saturday in February, one of Substack’s most widely read financial newsletters published a thought experiment: What if the AI boom, which has already minted extraordinary wealth and spurred corporate capital expenditure to historic highs, actually turns out to be a bearish signal instead of a bullish bubble waiting to burst? What if the same technology making white-collar workers more productive will soon destroy the larger white-collar economy?

The widely read Substack post by Citrini Research began with a hypothetical future memo dated June 30, 2028: “The unemployment rate printed 10.2% this morning, a 0.3% upside surprise. The market sold off 2% on the number, bringing the cumulative drawdown in the S&P to 38% from its October 2026 highs.”

In the real world present, Citrini’s thought experiment rocked the market. The Dow fell 1.7% that Monday. Individual stocks mentioned in the post — Monday.com, DoorDash — fell about 7% each. IBM fell almost 13%.

In other words, a Substack post laying out a theoretical scenario caused a real-world multibillion-dollar wipeout. And that may be an even more revealing read on the economy than the Citrini Research post that kicked it off. Would an implausible or far-fetched scenario have created that kind of reaction? Or did the post touch on very real, widespread, yet quiet fears — and lay bare how little choice any of us may have about the AI future? 

The white-collar contraction is already happening

In fact, the question Citrini posed — what happens to an economy built around the premium put on human intelligence when that premium disappears — is one that economists, labor market researchers, and workers themselves are increasingly asking. And while the data hasn’t fully come in yet, the early signals are striking.

White-collar payrolls have now contracted for 29 consecutive months. According to Aaron Terrazas, a former chief economist at Glassdoor, that’s without precedent. “It's clear that white-collar hiring has slowed and white-collar payrolls have contracted. This is incredibly unusual, going back 70, 80 years,” Terrazas said in an interview. “The fact is, we have not seen this long of a contraction in white-collar jobs outside of a recession ever before. That has to be kind of ringing some alarm bells.”

But the headline unemployment rate — still hovering around 4.3% — obscures this narrower white-collar issue. Terrazas argued that the number has become a less reliable signal than it once was, as labor market slack increasingly appears as underemployment and workforce exits rather than formal unemployment. The more telling indicators, he said, are job postings and hiring rates, both of which have been depressed for some time. “We're kind of getting smoke signals in all of these different corners of the economy right now,” he said.

Daniel Keum, a professor at Columbia Business School who studies AI in the workplace, is less circumspect. He said that AI is causing demand for white-collar workers to fall — no bones about it. He described the current moment as a “technological shock” with two distinct parts.

The first is already here: AI is replacing white-collar labor, not augmenting it, at least in the U.S.

“U.S. labor costs are very expensive,” Keum said in an interview. “So AI is targeted squarely at replacing people and reducing headcounts. That pays off handsomely.” 

The second part is AI creating a positive shock on the revenue side as it helps companies generate new products, services, and therefore new jobs. That dynamic is coming, Keum said,, but it could still be years away. Right now, we’re absorbing the negative cost shock without yet seeing the positive revenue shock.

Not all the displacement looks like a human being swapped out for an AI agent, Keum said. Some workers are losing jobs not because their specific roles have been automated, but because companies are reallocating resources toward AI and away from everything else. Meanwhile, those eye-popping corporate capital expenditure numbers — the hundreds of billions Amazon $AMZN, Microsoft $MSFT, Google $GOOGL, and Meta $META are pouring into AI infrastructure — are not translating into hiring, because they're going into data centers, not people.

Essentially, dramatic rises in business spending don’t equal demand for more workers with degrees, Keum argued. In fact, it may indicate the opposite.

As an indicator of demand for white-collar labor, Keum suggested looking to new MBA graduates. As arguably the most credentialed, most in-demand workers in the knowledge economy, their outcomes function as a leading indicator of elite labor demand. If companies are pulling back on even their most desirable hires, something has shifted.

And that data is not encouraging. In January, The Wall Street Journal reported that, at Duke $DUK University's Fuqua School of Business, 21% of job-seeking graduates were still looking for work three months after graduation last year — up from 5% in 2019. At Georgetown's McDonough School, that figure was 25% last year, up from 8% in 2019. At Michigan's Ross School, it was 15%, up from 4%. Even Harvard Business School still had 16% of graduates unemployed after three months, higher than before the pandemic.

To be clear, AI is unlikely to be the only force at work. Immigration policy shifts under President Donald Trump have complicated the pipeline for foreign graduates who once expected to secure U.S. work visas. Major tech firms are still digesting, and in some cases unwinding, the hiring sprees they went on after the pandemic. And elevated interest rates may have functioned to temper corporate exuberance. Volatile and even chaotic trade policy has rattled confidence from glass-walled c-suites to humble corner stores.

But even accounting for those factors, weakening demand for the most credentialed segments of the labor market stands out. If even the most elite business schools are sending a growing share of graduates into prolonged job searches, something fundamental has shifted.

The wage deflation piece is harder to see. But it’s there

The Citrini post likewise spent a lot of time unpacking the potential for white-collar wage deflation — a dynamic Keum is also watching. Traditionally, worker pay has closely been tied to productivity, he said. And AI is causing workers to become more productive. But because of falling demand for their labor, workers are having more difficulty capturing that extra value they create.

It’s simple: When some form of automation could substitute for your labor, your ability to negotiate is seriously weakened. “A junior associate at a law firm before could demand 20% of billable hours,” Keum said. “Now you bill more, but you take 10% — because if you demand anything more, there’s AI.”

If AI weakens labor’s ability to capture greater value, that could accelerate a longer-term trend. In the U.S., labor’s share of GDP — a proxy for how much value workers capture vs. how much capital captures — has been slowly declining for decades, falling almosty 10 percentage points from its peak in the late 1960s and early 1970s to 56% in 2024.

Still, it can be difficult to pin down whether white-collar compensation in particular is falling, in part because more granular data is hard to come by, and also because salaries can be sticky even as overall compensation may fall. From one year to the next, companies tend not to lower salaries directly because workers resist that; no one loves to see their paychecks actively shrink. However, companies can change the bargain in other ways that don’t necessarily announce themselves.

Terrazas, the former Glassdoor economist, described three buckets of other possible pay cuts. First, benefits packages may quietly shrink. For example, an employer may cover less of a health insurance premium than it once did. Second, non-salary compensation may become less generous, whether in the form of reduced stock grants or bonuses getting cut. Finally, the job itself may expand — duties increasing, hours lengthening — without any corresponding increase in pay.  The last is “kind of like shrinkflation,” Terrazas said, borrowing the consumer-pricing term for when a bag of chips gets smaller without the price changing.

All these factors potentially undercut compensation, even if salary numbers don’t change. And here again, there are signs that compensation is being cut in these ways: According to recent benefits data from Sequoia, the share of companies offering health plans that fully cover employee-only premiums has fallen for three straight years. 

While moving from zero-cost coverage to standard market-level cost sharing may not register in headline wage data, it reduces take-home compensation all the same.

A dark vision of cascading effects

Citrini’s hypothetical exercise offered a dark vision of white-collar layoffs and workers’ reduced earning power cascading through the larger economy — transforming prime mortgages into credit risks, shrinking the “demand base” for a swath of goods and services from cars to vacations and private schools. A “consumption hit” that would be “enormous relative to the number of jobs lost,” as Citrini put it.

Still, asked about this dark vision, Terrazas was clear: “To date, the evidence suggests modest rather than tectonic shifts, and there’s not yet a smoking gun that directly implicates AI — only a lot of smoke. Data is unavoidably backward-looking, so maybe it’s just a matter of time. The scenario described here would be outside the historical experience, but sometimes things really are different.”

“I think most people will agree that workers will adapt — are adapting — to these labor market shifts,” Terrazas said. “So the question then becomes: Will they be better or worse off after the adaption? The Citrini authors seem to assume that people will adapt for the worse — partial pivots like taking lower-paying or less prestigious work. I’m not sure that’s always, or has to be, the case.”

So not everyone agrees that alarm is warranted. Even some top economic officials pushed back on the Citrini post directly, with Federal Reserve Governor Christopher Waller saying that “AI is a tool. It's not going to replace us as human beings. This is kind of an overstated thing.” That view has history on its side. As Keum and Terrazas both noted, every prior wave of automation eventually created more jobs than it destroyed.

But the historical arguments depend on a key assumption: that whatever new jobs emerge will require humans to do them. That assumption, for the first time, is genuinely in question. Previous technologies — from the washing machine to the PC — eliminated specific tasks while human creativity and judgement remained the irreplaceable inputs. We can’t know that the future will look like the past.

The Citrini post arguably moved markets precisely because it so vividly portrayed that this time could be different. The optimistic view holds that white-collar workers will adapt, and that they'll ultimately land somewhere better. Maybe that will prove true. 

But if the present moment is any indication, the future being built doesn't seem to be one where white-collar workers have more power. It seems to be one where they have less.