The Hollow Middle

The middle class was never a natural feature of market economies. It was a political construction: built deliberately, through specific institutions, in a specific historical window. Progressive taxation. Public universities. Union contracts that tied wages to productivity. These were choices. And they produced something that had never existed before at scale: a broad stratum of people who could reasonably expect that education would be rewarded, that work would be stable, and that their children would do at least as well as they had.

That construction is now under strain from multiple directions simultaneously, including the persistent undervaluation of care and education work in a market that cannot price the public goods they generate, and the institutional mechanisms that built that hierarchy of “skill” in the first place as discussed in my previous piece. This piece asks a harder question. If we do not intervene, if we continue to let the current dynamics run: what does the economy look like in ten to fifteen years?

The answer: a slow, structurally reinforced disappearance of the middle.

What We Know Right Now

Several trends are already measurable and directional, not merely speculative.

Since 2019, the American economy has delivered by almost every headline measure. GDP expanded. Corporate earnings hit records. Even wages, on paper, went up. But adjust for inflation and the picture fractures. Real wages for American workers rose 4.5% between 2019 and 2025. Fortune 500 profits rose 65%. Nominal gains dissolved in the wash of rising prices, leaving the median worker with $16 more per week in actual purchasing power across six years.

The gap is 60.6 percentage points. Profits of the largest 500 companies grew 14 times (!) faster than real wages. That ratio is the exact arithmetic of who captured the expansion.

The labor share data makes the structural case. Workers’ share of output in the U.S. nonfarm business sector fell to around 53.8% in Q3 2025, near the lowest levels observed in the postwar data. The decline started before COVID, accelerated through the recovery, and continued even as GDP printed strong quarters.

Growth, in other words, is real. Its distribution is the problem. And that problem is not uniform. It lands differently depending on where in the labor market you stand. The aggregate numbers mask a split that has been widening for years.

The job market is bifurcating in a specific pattern. US employers added 584,000 jobs in 2025, the weakest year outside an official recession since 2003, down from 2 million the year before. But the collapse has not been evenly distributed. Entry-level professional hiring has fallen sharply: job postings for roles paying above $96,000 reached decade lows in early 2025, law firm entry-level hiring dropped 25% roughly, and 40% of white-collar job seekers in 2024 received zero interviews. Meanwhile, the occupations growing in largest volume are at the bottom of the wage distribution: home health aides, personal care, teachers, as a I argued in my analysis The Jobs AI Cannot Do Are the Jobs We Pay the Least. The middle is where the jobs are disappearing.

Consumer sentiment has registered this. Consumer sentiment has weakened materially. Consumer sentiments remained historically moderate in April 2026, amid war in Iran and economic uncertainties, with declines broad-based across groups, even as headline economic growth has remained resilient. Consumer spending growth is now driven overwhelmingly by the top 20% of income earners. The bottom 60% are drawing down savings and taking on credit to maintain standards of living they have not seen improve. The divergence between the aggregate headline and the lived middle-class experience is not a perception gap. It is a distributional reality that the headline obscures.

And then there is the credential trap. For a generation, the social contract in America was legible: borrow to get educated, the degree will pay off. That contract is now broken at both ends. The cost of a four-year degree has continued to rise even as the college wage premium has largely plateaued since the early 2000s, compressing the financial return to college rather than expanding it. The jobs that degree was supposed to unlock: the entry-level analyst, associate, junior consultant, first-year attorney, are precisely the roles most exposed to AI substitution. Young people are being told simultaneously that they need expensive credentials and that the jobs those credentials were supposed to produce are contracting.

The debt is real. The payoff is not.

The Scenarios

What follows is a structured exploration of how the trends documented above could resolve, depending on two genuine uncertainties:

whether AI-driven growth scales with or without broad labor input, and

whether institutions develop the capacity to shape how that growth is distributed.

These uncertainties are independent. Together they produce four distinct futures. The evidence currently available most closely supports the first.

The first uncertainty is technological and organizational: does AI function primarily as a labor-augmenting technology, sustaining demand for broad employment, or as a labor-automating technology, allowing firms to scale output without proportional increases in headcount? This is not yet settled. The current data with 584,000 jobs added in 2025 points toward labor-automating direction. But the direction is not irreversible, and it varies by sector.

The second uncertainty is institutional: do wage-setting mechanisms, labor organization, public investment, and job classification systems adapt to the new production structure, or do they lag it? Historically, institutions have lagged often by a generation.

Scenario 1: The Hollow Middle

(Labor-light scaling + weak institutions)

This is where the current data points.

Firms scale output without scaling headcount. Entry-level roles are not eliminated but compressed: fewer hires, higher thresholds, narrower pipelines. What had been a three-to-one junior-to-senior hiring ratio compresses toward one-to-one. Tasks once performed by junior workers are absorbed into AI-assisted workflows and redistributed upward. Fewer entry points mean fewer workers accumulate the experience required to transition into stable, middle-income roles. Over time, this reduces not just hiring but mobility itself.

Wage growth does not track productivity. The gap already visible in the data with Fortune 500 profits rising 65% against real wage growth of 4.5% between 2019 and 2025, widens as firms capture efficiency gains without expanding their workforce. Workers, increasingly fragmented across contract, platform, and hybrid arrangements, lack the leverage to claim a share of those gains. Labor’s share of output, already at 53.8% in late 2025 near its postwar low, continues to decline. These workers appear as employed in BLS measures. They do not appear in the white-collar unemployment statistics that frame policy debate.

Demand adjusts accordingly. Consumer spending, already driven overwhelmingly by the top 20% of earners, concentrates further. The bottom 60%, drawing down savings and taking on credit to maintain living standards, exhaust that buffer. The economy continues to grow — but the base of that growth narrows.

The care economy makes the endpoint legible. By the early 2030s, the largest-growing occupation in the United States by absolute count is home health and personal care work: 772,000 new positions projected between 2024 and 2034, at a median of $26,000 per year, with 36% of workers living in or near poverty. The new AI-adjacent roles are being priced by the same institutional mechanisms: not by task complexity, but by the bargaining power of the workers performing them. Which is to say: not much.

The system does not collapse. It stabilizes around a structure that no longer requires a broad middle class to function.

Scenario 2: The Engineered Middle

(Labor-light scaling + strong institutions)

The underlying production model is identical to Scenario 1. Firms still scale without proportional labor. Entry-level compression still occurs. The market, left to itself, still concentrates gains.

What changes is the response. As mobility weakens and cost-of-living pressures intensify, political pressure builds. Government intervenes in wage-setting, expand public investment in care and education, and establish minimum standards in fragmented labor markets. Intervention emerges unevenly across sectors and geographies, shaped as much by political constraints as by economic logic: some sectors see wage floors, others see firms accelerate automation to offset higher labor costs, others see costs passed to consumers.

The middle class persists, but the middle class is sustained, not generated. Its stability is contingent on the political coalitions that maintain it, which are themselves contingent on the very conditions the policy is trying to preserve. Less efficient by conventional measures, more stable by distributional ones. The trade-offs are explicit and permanent.

Scenario 3: The Competitive Diffusion Economy

(Labor-augmenting scaling + weak institutions)

Capability diffuses, lowering barriers to participation.AI reduces the cost of analysis, optimization, and coordination, allowing more firms and workers to perform high-value tasks without scale. Entry expands, and labor demand spreads rather than collapses. Competition partially distributes gains.As more actors enter, firms compete for talent capable of deploying AI effectively, limiting excess margins and pushing some gains into wages and prices. You can find more elaboration on the conditions of market creation using the energy sector as an example (in Mandarin though…) in Wendy’s podcast episode “Power to People”.

Though the key difference is technological, not the political intervention on labour, diffusion depends on market design. Data access, system interfaces, and regulatory structures must remain sufficiently open for new entrants to deploy capabilities; without this, diffusion stalls and re-concentration begins. However, control may re-concentrate even as participation expands.Platforms that coordinate demand, manage access, or control interfaces become new bottlenecks, shifting power from assets to systems rather than eliminating it.

As a result, the middle class persists, but becomes thinner and less stable.Workers benefit from broader participation, but without strong institutions, income gains remain uneven and reversible, producing a more open yet more volatile economic structure.

Scenario 4: The Reconfigured Middle

(Labor-augmenting scaling + strong institutions)

AI genuinely complements what workers do, creating new task combinations rather than simply removing existing ones. Institutions adapt: training systems, classification frameworks, wage-setting mechanisms, often lagging the pace of technological change but moving in the same direction. New middle-skill roles are actively constructed rather than assumed to emerge.

The middle class in this scenario is rebuilt differently: less linear career paths, more hybrid roles, less front-loaded credential debt. Transition costs are high and unevenly distributed. Some cohorts lose out. The system is more actively managed, more politically contested, and more fiscally demanding than the one it replaces. Admittedly, it is optimistic only in the sense that the middle class, having been deliberately constructed once, can in principle be deliberately constructed again.

The Point of Divergence

These scenarios diverge along observable pressures, each trackable in current data:

Whether firms continue scaling output without proportional labor

Whether AI capabilities commoditize or re-concentrate in dominant platforms across sectors

Whether workers retain or lose collective wage-setting power

Whether institutions adapt before or after these shifts become self-reinforcing

At present, the data aligns most closely with the Hollow Middle. But it does not guarantee it.

THE FORK

The trends described above do not point to a single outcome. They point to a system already under strain, moving in a direction that may become difficult to reverse. The real question is not which scenario is possible. It is when one path becomes structurally locked in.

That point emerges when three conditions, already visible in the data, begin to reinforce each other simultaneously:

Entry pathways narrow below replacement level, so fewer workers accumulate the experience that produces middle-income mobility.

Bargaining power erodes, as work fragments across platforms, contracts, and classifications too dispersed for collective leverage.

And consumption decouples from broad income growth, as high-income households sustain demand and firms serving that demand lose any market incentive to care about conditions further down the distribution.

Once these conditions persist across cycles through expansion, the system stops correcting itself: Growth no longer requires a broad middle class to generate it. Workers cannot collectively renegotiate their position. And political pressure to rebalance weakens as economic precarity distributes across millions of households, each experiencing the shift individually without a shared account of what is happening. This is what the hollow middle looks like as a new equilibrium, in which the middle persists as a social identity and a consumption aspiration long after it has ceased to be an income reality.

Historically, this kind of imbalance has been resolved, but only after it became legible. The institutional settlement of the Industrial Revolution, labor law, the welfare state, progressive taxation, for instance, arrived a generation after the growth that made it necessary. Can we afford wait that generation again?

The difference this time lies in its form. The effects are diffuse, distributed across balance sheets rather than concentrated in labor markets, visible in debt, housing, and retirement security rather than unemployment. Each adjustment is individually manageable. Together, they constitute a structural shift. Its speed and cumulative nature make it harder to recognize and respond to. The system absorbs its own consequences without producing the kind of crisis that forces recognition.

By the time the signal is unambiguous, the structure may already be set. What remains a question is whether the conditions that once built the middle class can be recreated before the system no longer demands one.

Cover photo by Marc Pell on Unsplash