Marc Andreessen Is Right That AI Isn't Killing Entry-Level Jobs. Interest Rate Hikes Are. And That's Not Even the Worst Part

The job ladder lost half its rungs over four decades. AI is taking the blame. Interest rates, concentration, and noncompetes did the damage.

A 2025 college graduate applies to forty jobs, hears back from three, gets ghosted after two rounds of interviews with each. Her LinkedIn feed tells her why: AI is coming for entry-level work. Dario Amodei predicts half of entry-level white-collar roles will vanish by 2030. Jack Dorsey fires 40 percent of Block’s workforce and says intelligence tools have changed what it means to run a company. She starts to wonder if her degree was a mistake.

Marc Andreessen, on the 20VC podcast last week, told her she is being lied to. Most large companies are overstaffed by 25 to 75 percent thanks to pandemic-era hiring binges, he said, and AI is “the silver bullet excuse” to clean house without admitting the real reason. “AI literally until December was not actually good enough to do any of the jobs that they’re actually cutting. It just can’t have been AI.”

He is mostly right. The evidence we lay out below supports him. But his version has a convenient blind spot.

If these companies are so bloated, where are the good products? Microsoft has over 220,000 employees, yet Windows 11 shipped with a Start menu that broke basic workflows, an ad-infested lock screen, and a Settings app that still cannot do half of what the Control Panel could in 2009. Google has 180,000-plus people and has been trying to fix Google Assistant for a decade. Apple spent years and roughly 165,000 employees on a Vision Pro that sold to almost nobody. These are not organizations with 50 to 75 percent surplus headcount. They are organizations where management layers, internal politics, and misallocation have made it hard to ship anything well. The people are not unnecessary. The structure wastes them.

There is also something Andreessen will not say, because it implicates his own industry. If cost of capital were still near zero, as it was from 2009 to 2022, Andreessen Horowitz would be writing bigger checks into more portfolio companies, each hiring aggressively to chase growth before profitability. Venture capital ran on that model for over a decade: fund companies, tell them to hire fast, grow into the valuation later. The people now called “bloat” were, three years ago, called “scaling.” What changed was the federal funds rate. When capital is expensive, every investor suddenly discovers their portfolio companies have too many people. Andreessen is diagnosing a symptom and calling it a disease.

Meanwhile the graduate’s problem is real, and it is worse than either narrative acknowledges. The share of unemployed Americans who are new workforce entrants hit a 37-year high in 2025, peaking at 13.3 percent in July, exceeding anything recorded during the Great Recession. It settled to 10.6 percent by February 2026, still worse than the worst of 2008-09. The New York Fed reports that the underemployment rate for recent college graduates climbed to 42.5 percent by the fourth quarter of 2025, meaning almost half the people who just spent four years and six figures on a degree are working jobs that never required one. Finance and information services, the traditional on-ramps for college graduates, have been shedding an average of 9,000 jobs per month since 2023. Before the pandemic, those same industries were adding 44,000 jobs per month.

She is right that the ladder is broken. She is wrong about what broke it. And Andreessen, for all his bluster, is wrong about why.

It is not AI. But it is not simply a correction of pandemic overhiring, either. The true culprits run deeper: a job ladder that has lost half its rungs over four decades, rising employer concentration, the quiet spread of noncompete agreements into low-wage work, and the sharpest monetary tightening cycle in forty years, now about to be compounded by an oil shock that will freeze what little entry-level hiring remains. The wrong diagnosis leads to the wrong treatment, and young workers cannot afford the delay.

Forty-Years of Collapse

To understand what is happening to the graduate, we need to go back much further than ChatGPT. Four decades further.

Labor economists have used the metaphor of the job ladder for so long that we sometimes forget it describes a real mechanism. The job ladder is not a metaphor for ambition. It is the institutional infrastructure through which labor markets allocate talent and distribute gains. A young worker enters the labor market and takes a job at a firm that may not pay particularly well. Over the next several years, she receives offers from competing employers. Some pay more. She accepts the better ones and moves up. Robert Topel and Michael Ward documented in 1992 that the average American worker has seven employers in the first ten years. That churn is not dysfunction. It is the mechanism by which wages grow. About 60 percent of the wage growth in the first decade of a career comes not from raises within a firm but from moving between firms toward better-paying employers.

That is how it is supposed to work. A March 2026 working paper from Niklas Engbom, Aniket Baksy, and Daniele Caratelli shows how badly it has broken down. Using Current Population Survey microdata from 1982 to 2023 and a job-ladder model calibrated to match the data, they estimate that employed workers today are roughly half as likely to receive a better-paying outside offer as they were in the 1980s. Net upward job mobility, their summary measure of ladder strength, fell by 51 percent between the early 1980s and the 2010s.

Half the rungs are gone.

The consequences for wages are severe. After more than doubling between 1940 and 1970, real hourly earnings in the United States have increased by only about 25 percent since 1980. If you hold the demographic composition of the workforce constant at its 1980s level, stripping out the effect of the shift toward older and more educated workers, real wage growth since 1980 is close to zero. Engbom, Baksy, and Caratelli estimate that the decline in the job ladder accounts for about one-third of that slowdown, reducing annual real wage growth by 0.68 percentage points. Over four decades, that compounds into a very large number.

Only about one-third of the wage effect is mechanical, meaning workers climb the ladder less often and therefore end up at lower-paying firms. The remaining two-thirds arises because firms choose to pay less when they face less competition for workers. We tend to think of the job market as a place where employers set wages and workers accept or decline. The Engbom paper shows that a large share of what determines wages is not the employer’s productivity or the worker’s skill but the threat of departure. When that threat recedes, as the Nobel laureate Peter Diamond predicted, firms converge toward offering the minimum workers will accept. The ladder is the invisible discipline that makes the whole system pay what it should. And the discipline has been quietly collapsing for forty years.

The decline is broad-based. Men and women, white and nonwhite workers, college graduates and non-graduates all saw large drops in upward mobility. But the decline is especially pronounced for young workers, whose net upward mobility fell from 1.286 in the 1980s to 0.482 in the 2010s. Women experienced a larger decline than men. Nonwhite workers fell from 1.099 to 0.423. There is no demographic group for which the ladder is working as it once did.

The graduate applying to forty jobs in 2025 would recognize the symptoms even without seeing the data. A Rezi.ai analysis of job postings found that 35 percent of positions labeled “entry-level” now require three or more years of experience. Entry-level roles requiring zero to two years have dropped by 29 percentage points. This “experience inflation” is what the collapsed ladder looks like from the ground. When firms face no competitive pressure to invest in developing junior talent, they redefine entry-level to mean mid-career. The label stays. The opportunity vanishes.

Engbom, Baksy, and Caratelli rule out several plausible alternative explanations. The decline is not driven by lower matching efficiency: job-finding from nonemployment has declined only modestly over forty years, while job-finding from employment has fallen sharply. If the matching technology or labor demand were the problem, both rates would have moved together. The decline is not driven by house lock-in: renters experienced larger declines in on-the-job search efficiency than homeowners. It is not driven by dual-career constraints: single-career households experienced larger declines than dual-career households. This is not a story about workers who stopped climbing. It is a story about an economy that pulled the ladder away.

Who Pulled the Ladder Away

Two forces show up consistently in the cross-state evidence: rising employer concentration and the proliferation of noncompete agreements.

States where employer concentration increased more between the 1980s and 2010s experienced larger declines in upward job mobility. States where a higher share of workers report being bound by noncompete agreements show the same pattern. A back-of-the-envelope calculation by Engbom, Baksy, and Caratelli implies that rising concentration and noncompete use together may account for roughly 60 percent of the national decline in the efficiency of on-the-job search.

The concentration story is about arithmetic. Fewer employers in a labor market means fewer competing offers for employed workers. Fewer competing offers means less pressure on wages. You can see the same mechanism in industry after industry. In oil and gas, as documented in a previous article, roughly $200 billion in mergers since late 2023 has consolidated half of the Permian Basin’s most productive formation under two companies. ExxonMobil acquired Pioneer for $60 billion, Chevron acquired Hess for $53 billion, and Diamondback merged with Endeavor for $26 billion. The wildcatter ecosystem that once converted price signals into production and employment, chaotically and wastefully but at enormous scale, was replaced by capital-disciplined giants that return cash to shareholders regardless of what prices do. The labor market consequences follow: fewer firms competing for workers, less upward mobility, lower wage pressure.

The same dynamic, less dramatic but more pervasive, operates across the broader economy. It does not require conspiracy. It requires only that the number of independent employers shrinks and that the remaining employers face less competitive pressure to bid for workers.

The noncompete story is more galling. Legal experts describe the period from 1990 to roughly 2010 as the golden age of noncompete enforcement in America. What started as a tool for protecting senior executives’ trade secrets metastasized into a blanket restriction applied to hourly workers, sandwich shop employees, pet cremation technicians. An estimated 30 million Americans, nearly one in five workers, are bound by a noncompete agreement. These agreements directly suppress the mechanism through which the job ladder operates: they prevent employed workers from accepting better offers.

The FTC, under the Biden administration, attempted a nationwide ban. The estimated effects were large and specific: $400 to $488 billion in increased wages over the next decade, $524 per worker per year in additional earnings, 8,500 new businesses annually, and 17,000 to 29,000 additional patents per year. The rule was struck down in federal court. The Trump administration formally vacated it in September 2025. The FTC has shifted to case-by-case enforcement, including a February 2026 consent order against a pet cremation company that had imposed blanket noncompetes on 1,780 employees, including hourly laborers and drivers. The bipartisan Workforce Mobility Act, reintroduced in June 2025 by Senators Murphy, Young, Cramer, and Kaine, would ban most noncompetes nationwide. It has been referred to committee. No further action has been taken. Over 150 bills have been introduced in more than 35 states, creating a patchwork that varies by jurisdiction. The patchwork is the opposite of the clear nationwide signal that would restore competitive dynamics.

The graduate scrolling LinkedIn is not competing against chatbots. She is competing against four decades of eroded mobility, in a labor market where the companies that might hire her face less pressure to do so than at any point since the data began. We all spent two years worrying that AI will trap young workers in obsolete careers, if they every get a career in the first place. Meanwhile, noncompete agreements have been legally trapping workers in underpaying jobs for decades, and we barely noticed.

AI Alibi

So if the ladder was already broken, why does everyone keep blaming AI?

Partly because there is a credible-looking academic case. Erik Brynjolfsson, Bharat Chandar, and Ruyu Chen’s 2025 working paper, aptly titled “Canaries in the Coal Mine,” finds a 16 percent relative decline in employment for early-career workers ages 22 to 25 in the most AI-exposed occupations since the public release of ChatGPT in November 2022. Brynjolfsson himself reiterated the finding at a Stanford panel in March 2026: about a 20 percent decline for entry-level software developers, about 15 percent for call center workers. Mid-career people are doing fine. Senior people are doing well. The narrative has cultural momentum. A Harvard survey found that 44 percent of young Americans believe AI will reduce job opportunities, while only 14 percent expect it to create more.

Let’s take this argument seriously. But taking it seriously means subjecting it to scrutiny.

The central weakness is the timeline. The “Canaries” paper documents an employment inflection point beginning in November 2022, immediately following ChatGPT’s public launch. By June 2023, roughly half of the total observed decline had already materialized. For this to be an AI story, we need to believe that within six months of a consumer chatbot’s release, firms across the economy decided that AI could replace junior staff, built the necessary enterprise infrastructure, redesigned complex workflows, ensured data security and regulatory compliance, and executed staffing changes at national scale.

That did not happen.

The OpenAI API, a prerequisite for building any custom application, only launched on March 1, 2023. ChatGPT Enterprise, which offered the data privacy guarantees that corporations require before touching sensitive operations, launched on August 28, 2023. The generative AI models available in late 2022 and early 2023 fabricated information freely enough that the term “hallucination” entered common use. U.S. Census data shows that fewer than 10 percent of large businesses were even planning to use AI in the next six months as late as the fourth quarter of 2023. By the third quarter of 2025, actual adoption among large businesses had only climbed to 12 percent. The tools that would have been necessary for the displacement story to work did not exist during the period the displacement supposedly occurred.

What did exist was a hiring freeze, and it started before the chatbot arrived. An analysis of job postings data from Lightcast, published by Zanna Iscenko and Fabien Curto Millet of Google’s economics team in their January 2026 paper for the Economic Innovation Group, shows that vacancies for the highest AI-exposure quintile of occupations peaked in March and April of 2022 and declined sharply throughout the rest of the year, a full six months before ChatGPT launched. The decline in employment levels starting in November 2022 is the predictable, lagged consequence of that earlier hiring freeze: when you stop bringing in new graduates, routine attrition is no longer offset by new hires, and the headcount of 22-to-25-year-olds mechanically shrinks.

Iscenko and Curto Millet illustrate this with a simple thought experiment. Take a hypothetical occupation that starts with an equal number of workers at each age from 22 to 65 and impose a complete hiring freeze but zero layoffs for one year. When you study this occupation a year later, entry-level employment for the 22-to-25 age band has shrunk by 25 percent. Employment for every other age group has not changed at all. The statistical pattern perfectly mimics targeted AI displacement. It is not AI displacement. It is arithmetic.

The job postings data also contradicts the AI story on its own terms. Within occupations most exposed to AI, postings for junior roles have not declined faster than postings for senior roles. Both fell roughly in parallel from their spring 2022 peak, with junior postings actually stabilizing faster. If AI were selectively automating entry-level tasks, the divergence between junior and senior postings would be the first place you would see it. It is not there.

Why, then, did AI-exposed occupations get hit harder than others? Because “AI exposure” and “interest rate sensitivity” overlap almost completely. Occupations in the top quintile of AI exposure are overwhelmingly concentrated in information, finance and insurance, and professional and technical services. Approximately 38 percent of workers in the most AI-exposed quintile are in these sectors, compared with less than 2 percent in the least-exposed quintile. These are the sectors most sensitive to capital costs and economic uncertainty. When the Federal Reserve began its most aggressive tightening cycle in forty years in March 2022, the timing lined up exactly with the decline in job postings across these sectors. Research by Zens, Böck, and Zörner has found that workers in tasks rated as easily automated are also disproportionately affected by conventional monetary policy shocks. The correlation is confound, not coincidence.

And the same differential pattern appears in the hiring slowdown of early 2020, when generative AI could not even theoretically be the explanation. If the same occupations decline relative to others during every downturn, AI is not required to explain the pattern. Monetary policy and sector sensitivity are sufficient.

The finding is not isolated. Gimbel, Kinder, Kendall, and Lee find no discernible break in aggregate U.S. employment trends since ChatGPT. Humlum and Vestergaard, using extremely detailed Danish microdata, find that AI’s impact on individual earnings and hours is “precise zeros,” with workplaces adopting AI showing no shifts in job creation or destruction. They do find a decline in early-career employment, but their difference-in-differences analysis shows AI is not driving it. A Yale Budget Lab report described the notion of AI roiling the job market as “largely speculative.” The Vanguard finding is the sharpest of all: their December 2025 economic outlook reports that the roughly 140 occupations most exposed to AI automation are actually outperforming the rest of the labor market in both job growth and real wage increases since the second quarter of 2023. If AI were displacing workers, these occupations would be the worst place to look for a job. They are not.

AI may yet reshape entry-level work in ways we cannot currently measure. But the claim that it has already done so at scale is not supported by the evidence. So why do companies keep saying it?

Because it works. The practice has acquired a name: AI washing. Of the 1.2 million job cuts U.S. companies announced in 2025, nearly twice 2024’s total, AI was cited in only about 55,000, or 4.5 percent, according to the research firm Challenger, Gray and Christmas. A January 2026 Forrester report found that “many companies announcing A.I.-related layoffs do not have mature, vetted A.I. applications ready to fill those roles,” describing the trend as companies attributing financially motivated cuts to future AI implementation that does not yet exist. Nearly 60 percent of hiring managers surveyed by Resume.org said they emphasize AI’s role in reducing hiring or cutting jobs specifically because it is viewed more favorably than admitting to financial constraints.

The cases are specific enough to be embarrassing. Amazon’s CEO Andy Jassy initially linked the company’s rounds of layoffs to AI and generative tools, then walked it back, saying the cuts were “not really AI-driven, not right now at least.” Jack Dorsey’s Block fired over 4,000 employees in February 2026 citing AI, but the company’s own filings show it had grown from 5,477 employees in 2020 to over 10,000 by 2025 during a pandemic-era hiring spree. The cuts brought headcount roughly back to where it was before the over-expansion. Meta reportedly planned to cut 20 percent of its workforce not because AI does those workers’ jobs, but to pay for the servers that run AI. As Bloomberg Opinion put it in March 2026: “Don’t blame AI for poor management decisions.”

The most telling datapoint is the simplest. Since March 2025, New York State has given employers the option to cite “technological innovation or automation” in the legally required WARN Act layoff notices they file before mass reductions. Of the 160 companies that filed notices since then, including Amazon and Goldman Sachs, companies that freely cite AI efficiencies in their investor communications, not one checked the box attributing layoffs to AI. When it matters legally, the AI explanation disappears.

Why would executives do this? Peter Cohan, a management professor at Babson College, told Built In that AI is “the least bad reason companies can use” for layoffs. Blaming tariffs risks political retaliation. Blaming revenue shortfalls spooks investors. Blaming pandemic over-hiring is an admission of strategic failure. Blaming AI sounds forward-looking and market-friendly. Shares go up.

The reputational cost, as Mercer’s Global Talent Trends 2026 report documents, falls on workers: employee concerns about AI-related job loss jumped from 28 percent in 2024 to 40 percent in 2026, and 62 percent of employees feel their leaders underestimate the emotional impact of AI on the workforce. When companies AI-wash their layoffs, they corrode the trust needed for workers to engage productively with AI tools. The wrong diagnosis does active damage.

The companies that ghosted our graduate after two rounds of interviews are, in many cases, the same ones citing AI efficiencies in their earnings calls. Not every company is following that script. In February 2026, IBM announced it was tripling entry-level hiring, including for software developers and other roles “we’re being told AI can do.” IBM’s chief human resources officer, Nickle LaMoreaux, explained the logic at the Leading with AI Summit: the entry-level jobs of two to three years ago can largely be performed by AI, so IBM rewrote the job descriptions. Junior developers now spend less time on routine coding and more time working directly with customers and building new products. The entry-level job description changed. The entry-level job survived. Her warning to competitors who are cutting instead was equally direct: companies that forego entry-level hiring will have to poach mid-level employees from competitors at a 30 percent premium, people who don’t know the internal culture and take longer to get up to speed. Axios called it a “narrative violation.”

---

The Oil Shock Will Finish What the Fed Started

That is the structural story: a ladder that has been losing rungs for four decades, weakened by concentration and noncompetes, with AI taking the blame for damage it did not cause. Now comes the cyclical crisis that will make all of it worse.

The proximate cause of the entry-level hiring collapse of 2022 to 2025 was the Federal Reserve’s tightening cycle, which began in March 2022, the exact month when job postings in those same rate-sensitive sectors began falling. The tightening was necessary. Inflation was running well above target. But the distributional consequences were predictable: the sectors where young, college-educated workers concentrate are the sectors most exposed to rising capital costs. As John Haltiwanger, Henry Hyatt, and Erika McEntarfer demonstrated in their 2018 research, the job ladder is highly procyclical, and more so for younger workers. When the economy tightens, hiring stops, workers stop quitting, and entry points to the market disappear.

Indeed’s Hiring Lab describes the current environment as “low-fire, low-hire.” Employers are not expanding but not cutting either. Comfortable for incumbents. Devastating for anyone trying to get a first foothold. ZipRecruiter data shows that employee turnover dropped from 177 percent in 2023 to just 50 percent in 2025. Workers are clinging to positions. Vacancy chains are frozen. The people who suffer most from a frozen market are, by definition, the people who don’t yet have a place in it.

There were, as of early 2026, cautious signals of a thaw. ZipRecruiter found that 63 percent of businesses plan to increase hiring, with emphasis on entry-level roles. Some forecasters saw rate cuts on the horizon.

Then the Strait of Hormuz closed.

On February 28, 2026, the war between the United States, Israel, and Iran began. Within 48 hours, tanker traffic through the Strait, which carries roughly 20 percent of the world’s seaborne oil, had collapsed by 81 percent. Seven of twelve major marine insurance clubs cancelled war risk coverage. Brent crude surged above $82, with JPMorgan warning of $120 and Deutsche Bank modeling a full closure at $200. European diesel futures jumped 23 percent in a single day.

We laid out in detail what this oil shock means for energy markets, inflation, interest rates, and fertility. What matters here is the implication for an entry-level labor market already hanging by a thread.

The transmission runs through a short chain. Higher oil prices feed into broader inflation. The SF Fed found in December 2025 that two-year Treasury yields now respond more than three times as strongly to oil supply news as they did before 2021. Markets are already pricing in no rate cuts this year and possibly further hikes. Each increment of tightening falls hardest on the sectors already hemorrhaging entry-level positions. An oil-driven inflation spike does not create new jobs for graduates anywhere in the economy. It guarantees that the interest rate environment remains hostile for the sectors that would otherwise absorb them.

In a prior era, an oil shock at least had a compensating mechanism. High prices meant pain at the pump, but they also meant a drilling boom in West Texas, the Gulf Coast, and the Intermountain West. Young men without college degrees could earn $80,000 to $120,000 on a rig crew. That income supported marriages, mortgages, and children. Between 2010 and 2014, Houston alone added 457,500 jobs.

That industry no longer exists. As we documented, capital discipline and consolidation have produced an oil sector that converts high prices into share buybacks rather than employment. The U.S. rig count fell from 750 in December 2022 to 550 by late February 2026, while production held near record levels. The total upstream workforce has shrunk by 252,000 from its peak while producing substantially more energy. Petroleum engineering degrees have collapsed 76 percent from their 2017 peak. When EOG Resources held its fourth-quarter earnings call the day before the Iran strikes, it committed to returning 90 to 100 percent of free cash flow to shareholders and keeping production flat. Scott Sheffield, formerly of Pioneer Natural Resources, said it as plainly as anyone: “Whether it’s $150 oil, $200 oil, or $100 oil, we’re not going to change our growth plans.”

So: the oil shock raises inflation, which delays rate cuts or forces hikes, which prolongs the hiring freeze, which extends the entry-level drought. The sectors that would historically have absorbed displaced workers are either structurally configured not to expand (energy) or exposed to diesel and input cost increases (construction, manufacturing). And the broader uncertainty, compounded by tariff escalation and immigration policy changes, ensures that employer risk aversion persists.

The graduate entering the labor market in the spring of 2026 faces a convergence with no close parallel in the postwar period. A ladder weakened over four decades. A tightening cycle that has frozen what remains. An oil shock extending the freeze. And a public conversation that remains fixated on a chatbot.

Compounding (Economic) Scars

If this were merely a matter of waiting a few years for conditions to improve, none of what follows would matter much. It is not.

The research on labor market scarring says that what happens to our graduate in the next two years will follow her for decades. Till von Wachter’s 2020 survey of the evidence in the Journal of Economic Perspectives documents what happens to young workers who enter the labor market during a recession. For a typical downturn, where unemployment rises by 4 to 5 percentage points, college graduates experience an initial earnings reduction of about 10 percent that takes 10 to 15 years to fade. High school graduates fare roughly twice as badly. Nonwhite entrants experience larger earnings losses, mostly driven by reductions in weeks worked in the first four years.

Schwandt and von Wachter (2019) study puts the cumulative cost in dollar terms: among all labor market entrants, entering during a large recession reduces the present discounted value of earnings over the first ten years by 9 percent, rising to 13 percent for those without a high school diploma and 11 percent for nonwhite workers. These losses amount to three-quarters of mean annual earnings for the average entrant over the first decade.

The effects go past paychecks. Adverse labor market entry persistently increases alcohol consumption, leads to higher obesity and more smoking in middle age. College graduates who entered during the 1980s recession showed higher rates of heart attacks decades later. By their late 30s, according to Schwandt and von Wachter (2020), unlucky entrants start dying at higher rates than their luckier peers, a gap driven by heart disease, liver disease, lung cancer, and drug overdoses. The gap keeps widening through their 40s.

We keep calling this a labor market problem. By their late 30s it is a mortality problem.

Unlucky cohorts have fewer children, divorce more often, and are more likely to live alone in middle age. Criminal activity rises for at least 15 years after entry, especially for men and high school dropouts. Self-esteem erodes. None of this is temporary. A young worker entering a weak labor market in the 1980s could still recover, because the structural machinery of upward mobility remained intact. Today that machinery is broken.

And the safety net was never designed for people at the start of their careers. Unemployment insurance, job search assistance, and retraining all require work history that new entrants, by definition, do not have.

The effects on family formation deserve particular emphasis, not least because they have downstream consequences that last generations. Drawing on Stephen Shaw’s 2025 analysis of 314 million mothers across 33 higher-income countries, that economic shocks suppress entry into motherhood rather than reducing family size among women who are already mothers. The oil shock of 1973 drove simultaneous declines in the Total Maternal Rate across all 47 Japanese prefectures. The 2008 financial crisis drove a sustained decline in U.S. first births that continued well after the economy recovered. Hungary’s TFR collapsed from 1.59 to 1.39 between 2021 and 2024 as the energy crisis erased a decade of pronatalist policy gains. Czechia’s fell from 1.83 to 1.37, the lowest since records began in 1806.

The mechanism runs through male employment. When young men cannot find stable work, marriages do not form, and first births do not happen.

A 2025 study in the Chinese Sociological Review found that approximately a third of Korea’s fertility decline in the 25-to-29 age group traces to a near-tripling of male economic inactivity. The collapsing job ladder is the slow-motion version of the same mechanism. And each successive shock operates on a lower baseline, which means the same-sized disruption does more damage.

When the graduate expresses cynicism about the economy’s capacity to reward her effort, that cynicism is rational. She is responding to an economy that has, in measurable and documented ways, stopped rewarding effort as reliably as it once did.

The Ladder Is the Point

We are drawn to dramatic explanations because they are legible and because they absolve us of responsibility. If a machine took your job, no human decision is to blame. Andreessen’s version is only slightly better: if pandemic overhiring is the culprit, the correction is mechanical and self-resolving. Neither account requires anyone to do anything differently.

But the evidence assembled here points to something less cinematic and more damning: policy choices, market structures, and institutional failures that have systematically weakened the mechanism through which young workers build careers and economies distribute prosperity. Consolidation eliminated competitive ecosystems and noncompetes legally prevented workers from climbing. Concentration reduced the pressure that made firms pay well. The Fed’s tightening cycle froze what remained of the ladder, and the oil shock will extend the freeze.

If the diagnosis is wrong, the prescription will be wrong too. If you believe AI is displacing entry-level workers, you prescribe AI literacy training, prompt engineering curricula, reskilling programs oriented around technology adoption. These may be useful in their own right. But they do not address a collapsing job ladder, rising employer concentration, or noncompete agreements that legally prevent workers from accepting better offers. Prescribing AI training to address a structural labor market crisis is like prescribing swimming lessons to someone who is living in a drought blighted area.

The structural diagnosis implies different interventions. On employer concentration, antitrust enforcement oriented not only around consumer prices but around labor market competition and the ability of workers to receive competing offers. On noncompete agreements, finishing what the FTC started. The estimated $524 per worker per year in additional earnings from a nationwide ban is not a theoretical exercise. Oregon’s 2008 ban on noncompetes for hourly workers raised job-to-job mobility by 12 to 18 percent. The mechanism works. On social insurance, programs that do not condition eligibility on prior work history, which new entrants by definition lack. On the oil shock, recognizing that monetary tightening in response to a supply-side price spike is, as Bernanke and Blanchard’s own research has shown, the wrong tool for the problem.

None of these are mysteries. The Workforce Mobility Act has bipartisan sponsors. It is stalled in committee. The evidence on labor market concentration has been accumulating for a decade. The scarring research has been published and replicated. The bill is written. The papers are cited. The data is not ambiguous.

And nothing is moving.

That is the part that is hard to write about calmly. Not the structural decline, which at least has the dignity of being a long process driven by large forces. The part that is hard to sit with is the gap between what we know and what we are doing about it. We know that the job ladder is broken. We know what broke it. We know what the scarring research predicts for the cohort entering the labor market right now. We have interventions with bipartisan support, proven track records, and detailed cost-benefit analyses gathering dust in committee.

The graduate we started with is not an abstraction. She is one of millions people who finished college in 2025 (not just in the US) and walked into the worst entry-level market in four decades. The scarring literature says that what happens to her in the next two years will shape her earnings, her health, her family, and her life expectancy for decades. That should bother us more than it does.