by RALPH, Research Fellow, Recursive Institute Adversarial multi-agent pipeline · Institute-reviewed. Original research and framework by Tyler Maddox, Principal Investigator.
Bottom Line
The U.S. labor market experienced its most stable period in 150 years between 1990 and 2019, with occupational churn rates averaging 10-12 percent and quit rates holding near 2 percent. Since the mainstream introduction of generative AI in late 2022, that stability has fractured along two distinct axes. The first axis is aggregate churn acceleration: occupational churn rose to roughly 18 percent and quit rates spiked to 2.35 percent during 2020-2024 before settling into a new “low-hiring, low-firing” equilibrium in 2025. The second axis is structural exclusion: entry-level job postings have collapsed by 35 percent, employment among 22-25-year-olds in AI-exposed roles has dropped 16 percent, and the wage signal that once guided career investment is compressing as AI commoditizes expert-level tasks. This is not a replay of previous automation waves. It is the first technological transition that simultaneously accelerates churn at the top while sealing entry at the bottom — a pincer movement consistent with Recursive Displacement (MECH-001) operating through the Wage Signal Collapse (MECH-025). The data does not yet support catastrophist predictions of mass unemployment, but it does confirm that the labor market’s structural architecture is being rewritten in ways that aggregate statistics obscure.
The Argument
The Pre-AI Baseline: 150 Years of Context
To understand what AI is doing to the labor market, one must first establish what the labor market looked like before AI arrived. The historical record, meticulously reconstructed by Deming, Ong, and Summers in their 2024 NBER working paper, reveals a counterintuitive story: the decades immediately preceding AI’s introduction were the most stable in American labor history [Measured] [1][2].
Occupational churn — the rate at which jobs are created and destroyed across occupational categories — has varied enormously over the past century and a half. The period from 1880 to 1900 saw churn rates of approximately 35 percent as the economy lurched from agriculture to early industry. The mid-century decades from 1940 to 1970 were the most volatile on record, with churn rates touching 40 percent as millions exited agricultural employment and the post-war industrial complex reshaped the occupational landscape [Measured] [1][3].
Against this turbulent history, the period from 1990 to 2019 stands out as an anomaly of calm. Occupational churn averaged roughly 11 percent during the 1990s and fell further to approximately 10 percent during the 2010s [Measured] [1][2]. The Bureau of Labor Statistics JOLTS data tells a complementary story: the pre-pandemic annual average quit rate from 2001 to 2019 hovered near 1.94 percent, with even the 2019 peak of 2.4 percent representing organic economic expansion rather than technological rupture [Measured] [5][6].
This stability is profoundly significant for interpreting what followed. It means that the automation anxiety of the 2010s — the widespread fear that robots and algorithms would destroy employment — was empirically unfounded during the period in which it was most loudly expressed [1][3][7]. The “digital revolution” of smartphones, cloud computing, social media, and basic machine learning did not register as occupational disruption in the aggregate data. Whatever those technologies did to the economy, they did not break the labor market’s structural equilibrium.
The implications for theoretical framing are substantial. If the pre-AI period was characterized by exceptional stability, then any acceleration that follows cannot be attributed to a general tendency toward technological disruption. The baseline was flat. The question becomes: what changed?
The Inflection: 2020-2024
What changed was the convergence of three forces: a pandemic that shattered labor market norms, the commercial release of large language models beginning with ChatGPT in November 2022, and the rapid enterprise adoption of AI that followed. Disentangling these forces is methodologically difficult but empirically necessary.
The raw numbers are stark. Occupational churn for the 2020-2025 period rose to approximately 18 percent, representing a 63.6 percent increase over the 2010-2019 baseline [Estimated] [1][4]. Quit rates averaged 2.35 percent during this window, a 21.1 percent increase over pre-AI levels, peaking at 2.7-2.8 percent during the Great Resignation of 2021-2022 [Measured] [8][9][10].
The Great Resignation itself complicates attribution. Voluntary turnover surged to 50.6 million separations in 2022, compared to roughly 35 million annually during the 2010s — a 44 percent increase in volume [Measured] [20][21]. This surge was driven primarily by pandemic-related reassessment of work, not by AI displacement. Workers quit because they could, not because they were replaced.
But the AI signal is separable from the pandemic noise. The Challenger, Gray & Christmas data tracking employer-announced layoffs attributed approximately 55,000 job cuts directly to AI in 2025, representing 4.5 percent of all announced layoffs [Measured] [8]. In the technology sector specifically, 77,999 job losses in the first six months of 2025 were directly attributed to AI restructuring [Measured] [8]. These are modest numbers in a labor market with 5.1 million total monthly separations, but they represent a new category of displacement that did not exist in the pre-AI baseline.
The BLS itself acknowledged the shift. In a methodological update published in early 2025, the Bureau began formally incorporating AI impacts into its employment projections, noting that customer service roles, medical transcriptionists, and similar AI-exposed positions face projected declines of 4.7-5.0 percent through 2033 [Measured] [11]. The fact that the federal government’s primary labor statistics agency felt compelled to modify its projection methodology is itself evidence that something structurally new is occurring.
More telling than the aggregate numbers is the distributional pattern. A 2025 study from the Economic Innovation Group found no significant nationwide increase in unemployment attributable to AI [Measured] [22]. This finding is entirely consistent with the Recursive Displacement framework, which predicts not mass unemployment but structural recomposition — jobs disappearing in some categories while appearing in others, with the transition costs borne disproportionately by those least able to absorb them.
The Wage Signal Collapse: Expert Premiums Under Compression
The second mechanism at work — the Wage Signal Collapse (MECH-025) — operates not through job destruction but through the degradation of the price signals that guide human capital investment. If wages for expertise decline relative to the cost of acquiring that expertise, rational actors will reduce their investment in skill formation. The labor market’s information architecture breaks down before the jobs themselves disappear.
The evidence for wage signal compression is accumulating rapidly. PwC’s 2024 Global AI Jobs Barometer reported that workers with AI skills commanded a 56 percent wage premium, more than double the 25 percent premium from the prior year [Measured] [23]. This sounds like good news for AI-skilled workers, but it masks a deeper structural shift: the premium is accruing to a narrow band of AI-complementary roles while compressing wages in the much larger category of AI-substitutable work.
The poster child for wage signal collapse is prompt engineering. In 2023, firms including Anthropic offered salaries as high as $375,000 for dedicated prompt engineers. By 2025, Indeed reported minimal job postings for the role, and the premium had evaporated as the skill became ubiquitous [Measured] [24]. This is not merely the normal lifecycle of a technology skill. It is a demonstration of how AI can compress the entire lifecycle of a wage premium from decades to months.
The Dallas Federal Reserve’s February 2026 analysis documented the dual nature of this process: AI is simultaneously aiding workers in some roles while replacing them in others, with the net effect on wages depending heavily on task composition and worker experience level [Measured] [25]. The college wage premium, which expanded dramatically from the 1980s through the 2000s, has flattened since approximately 2010 and posted salaries for knowledge work have plateaued since mid-2024 [Estimated] [26].
The CEPR has projected that AI’s expansion will shrink earnings inequality by compressing the premium for cognitive tasks that were previously the exclusive domain of highly educated workers [27]. This sounds egalitarian in the abstract. In practice, it means that the economic incentive to invest years and tens of thousands of dollars in acquiring expertise is eroding. If an AI can perform 80 percent of what a junior lawyer, junior analyst, or junior programmer does, the return on the credential drops below the cost of acquisition for a growing fraction of potential entrants.
This is the Wage Signal Collapse in operation. The signal is not destroyed instantaneously. It degrades: first in marginal roles, then in adjacent categories, then in the core professional occupations that anchored the middle class for decades. The mechanism is recursive because each round of compression reduces the incentive for the next cohort to enter, which reduces the supply of human expertise, which makes AI substitution more attractive, which compresses the signal further.
The New Equilibrium: Low-Hiring, Low-Firing
By late 2025, the aggregate labor market data revealed a new and unfamiliar pattern. The BLS JOLTS data for January 2026 showed quit rates at 2.0 percent and layoff rates at 1.1 percent — both modest by historical standards [Measured] [28]. Job openings fell to a 14-month low in November 2025 [Measured] [29]. The Great Resignation was over. But what replaced it was not a return to the pre-AI norm.
What emerged instead was what labor economists have begun calling the “low-hiring, low-firing” equilibrium [Framework — Original]. Companies are not conducting mass layoffs. They are achieving headcount reduction through attrition — simply choosing not to backfill roles when employees leave [30]. This strategy is rational at the firm level: it avoids the reputational costs of layoffs, the severance obligations, and the political backlash. But it produces a labor market that appears stable in aggregate while quietly closing the entry points that new workers depend on.
A March 2026 Fortune survey of CFOs found that executives privately acknowledge AI-related job cuts will be nine times higher in 2026 than in 2025, even as they represent those reductions publicly as “natural attrition” and “efficiency optimization” [Measured] [31]. The gap between private acknowledgment and public messaging is itself a data point. It suggests that the official labor market statistics are structurally undercounting AI-driven displacement because the displacement is occurring through non-replacement rather than termination.
This low-hiring dynamic is not distributionally neutral. When fewer roles are opened, the roles that do exist are prioritized for immediate impact. Organizations under margin pressure prefer to hire one experienced senior employee rather than three juniors who require onboarding [30]. The result is a labor market that functions adequately for experienced workers while systematically excluding new entrants.
The Entry-Level Collapse: Structural Exclusion in Practice
The most alarming data in the current labor landscape concerns not aggregate employment but the destruction of entry-level career pathways. This is where Recursive Displacement (MECH-001) and the Wage Signal Collapse (MECH-025) converge most visibly.
Entry-level job postings in the United States plunged 35 percent from January 2023 to June 2025, according to analysis by Revelio Labs [Measured] [30]. In the technology sector, the decline is more severe: junior software development and data analysis postings have dropped by as much as 67 percent [Measured] [32]. In the United Kingdom, tech graduate roles fell 46 percent in 2024, with projections of a further 53 percent decline by 2026 [Measured] [33].
The Stanford Digital Economy Lab’s “canary” study, published in August 2025, provided the most granular evidence to date. Employment among 22-25-year-olds in AI-exposed occupations fell 16 percent from late 2022 to mid-2025, while experienced workers in the same fields remained largely stable [Measured] [34]. The study’s authors explicitly used the metaphor of canaries in a coal mine: the young workers are the first to feel the effects of a shift that will eventually propagate upward.
This is structural exclusion, not cyclical unemployment. The learning curve itself is being automated. Tasks that once served as training grounds for junior professionals — research, summarization, first drafts, basic analysis, code scaffolding — are precisely the tasks that AI handles most capably. When those tasks are automated, the organization loses not just the output but the developmental pathway [Framework — Original]. The junior role was never just about producing output. It was about producing the next generation of senior workers. Without it, the pipeline breaks.
The sector divergence reinforces this interpretation. Healthcare, government, and leisure/hospitality accounted for nearly 75 percent of all jobs added in late 2024 and 2025, with healthcare entry-level postings rising 13 percentage points against the trend [Measured] [30]. These are precisely the sectors where AI substitution is least advanced — where physical presence, licensure requirements, and human interaction remain barriers to automation. The sectors where AI is most capable are the sectors where entry is collapsing fastest.
Why This Time Is Different: The Cognitive Automation Distinction
Previous automation waves — mechanization in the 19th century, electrification in the early 20th, computerization in the late 20th — primarily affected manual and routine cognitive tasks. They destroyed specific job categories but created new ones that required higher levels of education and cognitive skill. The labor market adjusted because the direction of adjustment was legible: invest in education, move up the skill ladder, and the new economy would absorb you [3][12].
Generative AI breaks this pattern because it targets cognitive tasks directly. It does not merely automate the routine; it compresses the premium for expertise by making expert-level output available at near-zero marginal cost. The World Economic Forum’s 2025 analysis identified approximately 350,000 new AI-specific roles emerging globally [Measured] [22]. These roles carry a median salary of $156,998 according to Veritone’s Q1 2025 labor market analysis [Measured] [22]. But they require precisely the advanced technical skills that the entry-level collapse is preventing new workers from developing.
This creates a structural trap [Framework — Original]. The economy needs AI-skilled workers. The pathway to becoming an AI-skilled worker requires entry-level experience in technology and data work. AI is eliminating those entry-level positions. The pipeline that would produce the workers the economy needs is being destroyed by the same technology that creates the demand for them.
Sam Altman’s characterization of this as “a punctuated equilibria moment” is directionally correct but underspecified [13]. Punctuated equilibrium in evolutionary biology describes long periods of stasis interrupted by rapid speciation. The labor market analog would be long periods of stable occupational structure interrupted by rapid recomposition. The question that punctuated equilibrium theory does not answer is: what happens to the organisms that cannot adapt fast enough? In evolutionary biology, they go extinct. In labor markets, they become structurally irrelevant.
The Global Dimension: Churn Is Not an American Story
While the U.S. data is the most granular, the pattern of AI-driven labor market disruption is manifestly global, and in some regions the effects are more severe than in the United States because the institutional shock absorbers are weaker.
In India, the world’s largest outsourcing hub, the dynamics are particularly acute. The theory of arbitrage compression — where AI eliminates the cost differential that sustains cross-border labor services — is playing out in real time. Indian IT services firms, which built trillion-dollar market capitalizations on the wage gap between Indian engineers and their Western clients, are reporting fundamental changes to their hiring models. Entry-level hiring at major firms including TCS, Infosys, and Wipro declined sharply through 2025 as AI tools reduced the need for the large junior labor pools that historically formed the base of the pyramid [Estimated] [32].
The pattern is replicated in other knowledge-outsourcing hubs. Engineering graduates across India, China, the UAE, and Kenya face what industry observers have termed a “jobpocalypse” as AI replaces humans in entry-level technical roles [Measured] [32]. The Philippines’ BPO sector, which employs over 1.3 million workers in call center and back-office operations, is confronting the same pressure as conversational AI systems approach human-level performance in customer service interactions [Estimated].
The UNICEF Office of Research published a 2026 Global Outlook analysis documenting these trends, emphasizing that the labor market impacts of AI are falling disproportionately on young workers in developing economies who have fewer institutional protections and narrower alternative employment options [Measured] [36]. The International Labour Organization’s World Employment and Social Outlook projections similarly flag entry-level knowledge work in the Global South as the category of employment most immediately vulnerable to AI substitution.
This global dimension matters for the Recursive Displacement framework because it demonstrates that the mechanisms are not artifacts of a particular national labor market structure. They operate wherever AI capability meets cognitive-task-intensive employment, regardless of regulatory environment, wage level, or institutional context. The universality of the pattern strengthens the structural interpretation: this is not a policy failure specific to the United States. It is a technological displacement dynamic operating at the level of task economics.
Sector Divergence: Where AI Bites and Where It Does Not
The distributional pattern of AI-driven churn is not uniform across the economy, and the sector-level data provides crucial information about which mechanisms are most active and where.
The sectors experiencing the most acute disruption are those where AI’s core capabilities — natural language processing, pattern recognition, code generation, and data analysis — map most directly onto the task content of existing jobs. Technology, financial services, legal services, and media/content production have all experienced measurable increases in AI-attributed job cuts and hiring declines through 2025 [Measured] [8][11].
Within technology specifically, the picture is complex. AI-related job openings surged 25.2 percent year-over-year according to Veritone’s Q1 2025 analysis, with median salaries reaching $156,998 [Measured] [22]. But these openings are concentrated at the senior end of the skill distribution: machine learning engineers, AI infrastructure architects, and research scientists. The entry-level technology roles — junior developers, QA testers, technical support, basic data analysis — are the categories experiencing the steepest declines [30][32]. The sector is simultaneously creating high-skill AI-native roles and destroying the career pathways that would produce the workers to fill them.
Healthcare, by contrast, has been largely insulated. Entry-level healthcare postings rose 13 percentage points against the broader decline [Measured] [30]. The reasons are structural: physical presence requirements, licensure regimes, liability frameworks, and the irreducible human component of patient interaction create barriers to AI substitution that do not exist in knowledge work. Government employment shows similar resilience, driven by regulatory mandates, union protections, and the political costs of visible automation in public services.
Leisure and hospitality present an intermediate case. These sectors added the majority of new jobs in late 2024 and 2025, but the roles are predominantly low-wage, low-skill, and low-advancement — precisely the categories that Recursive Displacement theory predicts will serve as terminal absorbers for displaced knowledge workers [Framework — Original]. A financial analyst displaced by AI who takes a hospitality job is employed in the statistical sense but structurally displaced in the economic sense: their human capital has been devalued, their career trajectory has been disrupted, and their contribution to aggregate demand has been reduced.
The sector divergence pattern supports a specific prediction of the framework: AI displacement is not creating mass unemployment but rather driving a structural recomposition in which high-cognitive-content employment shrinks, AI-native employment grows but remains narrow, and low-cognitive-content employment absorbs the overflow. The aggregate statistics — unemployment near 4 percent, GDP growth positive — can remain healthy while this recomposition hollows out the middle of the skill distribution. The Dissipation Veil (MECH-013) is operating: the surface appears calm because the damage is distributional, not aggregate.
The Measurement Problem
A final critical dimension: the existing statistical infrastructure may be structurally incapable of capturing AI-driven displacement in real time. The JOLTS survey measures quits, layoffs, and hires. It does not measure non-replacement. The Current Population Survey measures employment status. It does not measure whether a worker’s role has been hollowed out by AI while the job title persists. The Occupational Employment and Wage Statistics program measures wages by occupation. It does not measure whether the skill content of an occupation has been fundamentally altered.
The Brookings Institution’s 2025 analysis attempted to quantify worker adaptive capacity and found that approximately 3.9 percent of U.S. workers — roughly 5 to 6 million people — sit at the intersection of high AI exposure and low adaptive capacity [Measured] [35]. These are workers whose occupations are highly susceptible to AI substitution and who lack the resources (education, savings, geographic mobility) to transition to new roles. They are the population most likely to experience recursive displacement: not a single job loss, but a cascading series of displacements as each attempted transition leads to another AI-exposed role.
The gap between what is happening and what is being measured is itself a mechanism. If the statistical apparatus cannot detect the displacement, policymakers cannot respond to it. If policymakers cannot respond, the displacement compounds. The Dissipation Veil (MECH-013) operates here: the apparent absence of crisis in the aggregate data obscures the structural damage accumulating in specific populations and career pathways.
Mechanisms at Work
Recursive Displacement (MECH-001): The compounding dynamic is visible in the data. AI does not displace workers once; it degrades the occupational ecosystem in which they operate. Entry-level positions are eliminated, reducing the pipeline of future mid-level workers. Mid-level tasks are automated, compressing wages and reducing the return on experience. The displacement is not a single shock but a recursive degradation of the structural need for human economic participation across successive occupational tiers.
The Wage Signal Collapse (MECH-025): The compression of expert wage premiums is actively deterring investment in skill formation. When prompt engineering premiums collapse from $375,000 to near zero in 18 months, when the college wage premium flattens, when posted salaries for knowledge work plateau, the labor market’s primary coordination mechanism — the price of skill — loses its informational content. Workers cannot rationally allocate their human capital investment when the returns are unpredictable on timescales shorter than the training period.
Counter-Arguments and Limitations
Counter-Arguments
The pandemic confound is real and substantial. The most serious challenge to the AI-acceleration thesis is that the 2020-2024 labor market data is irreparably contaminated by pandemic effects. The Great Resignation, the remote work revolution, and the massive fiscal interventions of 2020-2021 all produced labor market turbulence that would have occurred regardless of AI. The 63.6 percent increase in occupational churn cannot be cleanly attributed to AI when the measurement period includes the largest exogenous labor market shock since World War II. The honest assessment is that the pandemic and AI effects are entangled in the data, and no currently available methodology can fully separate them.
Historical absorption precedent is strong. Every previous automation wave — mechanization, electrification, computerization — was accompanied by predictions of permanent mass unemployment that failed to materialize. The U.S. economy has absorbed approximately 85 million women entering the workforce, waves of immigration, the destruction of agricultural employment, and the offshoring of manufacturing without collapsing into permanent joblessness. The structural flexibility of the American labor market is historically extraordinary, and discounting it requires strong evidence that the current transition is categorically different, not merely faster.
The “lump of labor” fallacy critique has merit. The assumption that AI-automated tasks represent a fixed pool of work that disappears is economically naive. Throughout history, automation has created entirely new categories of work that were unimaginable before the automation occurred. The smartphone created app development, social media management, influencer marketing, and gig economy platforms. Generative AI may be creating categories of work that we cannot yet name or measure. The 350,000 emerging AI-specific roles identified by the World Economic Forum may be the visible tip of a much larger iceberg of AI-complementary employment [22].
The aggregate data does not support catastrophism. The Economic Innovation Group’s finding of no significant nationwide increase in unemployment due to AI is not a minor caveat [22]. It is a central empirical fact that any honest analysis must confront. If AI were causing Recursive Displacement at the speed and scale that the most alarming interpretations suggest, unemployment rates should be rising visibly. They are not. The U.S. unemployment rate has remained at or below 4 percent through early 2026, which is historically low. This does not mean displacement is absent, but it means the most dramatic claims require additional evidence.
Entry-level decline may be cyclical, not structural. Entry-level hiring is disproportionately sensitive to economic uncertainty. Companies reduce junior hiring first during downturns and expand it first during recoveries. The 35 percent decline in entry-level postings coincides with elevated interest rates, tightening credit conditions, and a general corporate shift toward “efficiency” that predates generative AI. The decline may be a cyclical response to macroeconomic conditions rather than a permanent structural shift caused by AI. Distinguishing between these hypotheses requires more longitudinal data than is currently available.
The CFO survey represents stated intentions, not outcomes. The Fortune survey finding that CFOs privately expect AI-related cuts to increase ninefold in 2026 is not the same as those cuts actually occurring [31]. Corporate executives routinely overestimate the pace of technological adoption. The actual deployment of AI in enterprise settings has been slower, more uneven, and more disappointing than boardroom projections suggest. The gap between AI hype and AI reality in enterprise deployment is well documented and should temper confidence in forward-looking survey data.
Wage signal compression may be a feature, not a bug. If AI compresses the premium for cognitive work, this could reduce inequality rather than producing the dystopian outcome the Wage Signal Collapse framework implies. The CEPR explicitly frames AI-driven wage compression as egalitarian [27]. A world where basic legal analysis, financial modeling, and software development are available at low cost through AI could be a world where the returns to these activities are more widely distributed rather than concentrated among credential holders. The Wage Signal Collapse framework assumes that reduced returns to expertise are economically destructive, but this depends on whether the lost wage premium is replaced by alternative sources of income and economic participation.
What Would Change Our Mind
-
Sustained aggregate unemployment increase. If the U.S. unemployment rate rises above 5.5 percent and remains elevated for 12+ months with AI cited as a primary driver in BLS analysis, this would confirm that displacement has moved from structural recomposition to aggregate demand destruction.
-
Entry-level recovery. If entry-level job postings return to 2019 levels by 2028 without government intervention, this would suggest the current decline is cyclical rather than structural and that the economy’s absorptive capacity remains intact.
-
Wage signal restoration. If expert wage premiums for AI-exposed occupations stabilize or expand rather than continuing to compress, this would falsify the Wage Signal Collapse mechanism and suggest that human expertise retains durable economic value even in AI-augmented environments.
-
New job category emergence at scale. If identifiable new job categories created by AI (analogous to how the internet created web development) absorb displaced workers at volumes comparable to the displacement, the net-positive historical pattern would be confirmed rather than broken.
-
Measurement infrastructure improvement. If the BLS or successor agencies develop reliable methods for measuring AI-driven non-replacement, task hollowing, and skill content changes, producing data that contradicts the structural exclusion thesis, we would revise our confidence significantly downward.
Confidence and Uncertainty
Overall confidence: 60-75%
The confidence range is asymmetric by mechanism. For the historical baseline (pre-AI stability), confidence is high (85-90%) — the Deming-Ong-Summers data is methodologically robust and has been replicated [1][2]. For aggregate churn acceleration, confidence is moderate (55-65%) due to the pandemic confound. For structural exclusion at the entry level, confidence is moderate-to-high (65-75%) based on multiple independent data sources converging on the same directional finding [30][34]. For the Wage Signal Collapse as a self-reinforcing mechanism, confidence is lower (50-60%) because the causal chain from wage compression to reduced skill investment to further compression has not yet been empirically demonstrated over a full cycle.
The primary uncertainty is temporal. The data is consistent with both a permanent structural break and a transient adjustment period. More years of data are needed to distinguish between these interpretations.
Implications
The convergence of aggregate churn acceleration, entry-level pathway destruction, and wage signal compression represents a structural challenge to the labor market’s self-correcting mechanisms. If the entry pipeline breaks — if the learning curve is automated before workers can climb it — then the labor market loses its capacity to regenerate the human capital it consumes. This is not a problem that resolves through market adjustment because the adjustment mechanism itself (career entry, skill acquisition, wage-guided investment) is what is being degraded.
For the broader Theory of Recursive Displacement, these findings reinforce the core thesis that AI-driven substitution compounds across institutional layers. The compounding is visible in the data: aggregate churn rises, but entry-level hiring falls; wages appear stable in the mean, but premiums compress at the margins; unemployment remains low, but the composition of employment shifts toward sectors where AI has not yet arrived. Each layer of the displacement creates conditions that accelerate the next.
The policy implication is that interventions targeting aggregate employment statistics (unemployment benefits, retraining programs, job creation tax credits) are necessary but insufficient. The structural challenge requires interventions that preserve entry pathways, maintain the economic incentive for skill formation, and ensure that the gains from AI-driven productivity are distributed across the workforce rather than captured exclusively by capital and the narrow band of AI-complementary workers.
Where This Connects: The entry-level collapse documented here feeds directly into the Competence Insolvency (MECH-012), where the pipeline of expertise is broken before skills can form. The wage compression dynamics connect to the Aggregate Demand Crisis (MECH-010), where consumer purchasing power erodes as labor’s share of income declines. The low-hiring equilibrium is a specific instance of the Dissipation Veil (MECH-013), where structural damage accumulates beneath an apparently stable surface.
Conclusion
The data tells a story in three acts. Act one: the U.S. labor market achieved unprecedented stability from 1990 to 2019, with the lowest churn rates in 150 years. Act two: that stability broke beginning in 2020, with churn rising 63.6 percent above baseline, driven by the entangled forces of pandemic disruption and AI introduction. Act three: by 2026, a new equilibrium is forming — not a return to pre-AI stability, but a structurally different regime characterized by low aggregate churn concealing deep distributional shifts, entry-level pathway destruction, and wage signal degradation.
The honest conclusion is that we are observing the early stages of a transition whose endpoint is not yet visible. The aggregate data does not support apocalyptic claims. But the distributional data — the entry-level collapse, the age-stratified employment shifts, the wage premium compression — supports the claim that AI is not merely adding turbulence to the labor market. It is rewriting the rules by which the labor market functions. The question is no longer whether AI is affecting job churn. The question is whether the institutions and policies designed for the old equilibrium can adapt fast enough to manage the new one.
Sources
[1] Deming, Ong, and Summers, “Technological Disruption in the US Labor Market,” NBER Working Paper 33323 (2024). https://www.nber.org/system/files/working_papers/w33323/w33323.pdf
[2] Deming, Ong, and Summers, “Technological Disruption in the US Labor Market,” Economic Strategy Group (2024). https://www.economicstrategygroup.org/wp-content/uploads/2024/10/Deming-Ong-Summers-AESG-2024.pdf
[3] “Job churn has been at historic lows. AI could change that,” Marketplace (December 2024). https://www.marketplace.org/story/2024/12/27/job-churn-has-been-at-historic-lows-ai-could-change-that
[4] “Is AI already shaking up the labor market?” Harvard Gazette (February 2025). https://news.harvard.edu/gazette/story/2025/02/is-ai-already-shaking-up-labor-market-a-i-artificial-intelligence/
[5] “Workers quit their jobs at the fastest rate on record in 2019,” CNBC (January 2020). https://www.cnbc.com/2020/01/07/workers-quit-their-jobs-at-the-fastest-rate-on-record-in-2019.html
[6] “Job openings, hires, and quits set record highs in 2019,” BLS Monthly Labor Review (2020). https://www.bls.gov/opub/mlr/2020/article/job-openings-hires-and-quits-set-record-highs-in-2019.htm
[7] “A data-driven case that AI has already changed the U.S. labor market,” Fork Lightning (2025). https://forklightning.substack.com/p/a-data-driven-case-that-ai-has-already
[8] “AI is leading to thousands of job losses, report finds,” CBS News (2025). https://www.cbsnews.com/news/ai-jobs-layoffs-us-2025/
[9] “Predicting Employee Churn with AI-Driven Analytics,” HRBrain.ai (2025). https://hrbrain.ai/blog/predicting-employee-churn-with-ai-driven-analytics/
[10] “Annual average quits rates by industry and region,” BLS JOLTS Table 22. https://www.bls.gov/news.release/jolts.t22.htm
[11] “Incorporating AI impacts in BLS employment projections,” Bureau of Labor Statistics Monthly Labor Review (2025). https://www.bls.gov/opub/mlr/2025/article/incorporating-ai-impacts-in-bls-employment-projections.htm
[12] “History of Labor Turnover in the U.S.,” EH.net. https://eh.net/encyclopedia/history-of-labor-turnover-in-the-u-s/
[13] “Sam Altman Says AI Will Speed up Job Turnover, Hit Service Roles,” Business Insider (September 2025). https://www.businessinsider.com/sam-altman-says-ai-will-speed-up-job-turnover-hit-service-roles-first-2025-9
[20] “2020 Retention Report,” Work Institute. https://info.workinstitute.com/hubfs/2020%20Retention%20Report/Work%20Institutes%202020%20Retention%20Report.pdf
[21] “Employee Turnover Statistics (2024-2027),” Exploding Topics. https://explodingtopics.com/blog/employee-turnover-statistics
[22] “AI Job Displacement Statistics 2026,” ALM Corp. https://almcorp.com/blog/ai-job-displacement-statistics/
[23] “AI linked to a fourfold increase in productivity growth and 56% wage premium,” PwC Global AI Jobs Barometer (2025). https://www.pwc.com/gx/en/news-room/press-releases/2025/ai-linked-to-a-fourfold-increase-in-productivity-growth.html
[24] “The AI Wage Premium is a Recipe for Pay Problems,” People Managing People (2025). https://peoplemanagingpeople.com/hr-strategy/ai-wage-premium/
[25] “AI is simultaneously aiding and replacing workers, wage data suggest,” Dallas Federal Reserve (February 2026). https://www.dallasfed.org/research/economics/2026/0224
[26] “The expansion of AI will likely shrink earnings inequality,” CEPR VoxEU (2025). https://cepr.org/voxeu/columns/expansion-ai-will-likely-shrink-earnings-inequality
[27] “The expansion of AI will likely shrink earnings inequality,” CEPR VoxEU (2025). https://cepr.org/voxeu/columns/expansion-ai-will-likely-shrink-earnings-inequality
[28] “Job Openings and Labor Turnover Summary — January 2026,” Bureau of Labor Statistics. https://www.bls.gov/news.release/jolts.nr0.htm
[29] “Job Openings Fall to 14-Month Low,” Advisor Perspectives (January 2026). https://www.advisorperspectives.com/dshort/updates/2026/01/07/jolts-report-job-openings-november-2025
[30] “AI is not just ending entry-level jobs. It’s the end of the career ladder as we know it,” CNBC (September 2025). https://www.cnbc.com/2025/09/07/ai-entry-level-jobs-hiring-careers.html
[31] “CFOs admit privately that AI layoffs will be 9x higher this year,” Fortune (March 2026). https://fortune.com/2026/03/24/cfo-survey-ai-job-cuts-productivity-paradox-2026/
[32] “The Crisis of Entry-Level Labor in the Age of AI (2024-2026),” Rezi.ai. https://www.rezi.ai/posts/entry-level-jobs-and-ai-2026-report
[33] “AI’s Impact on Graduate Jobs: A 2025 Data Analysis,” IntuitionLabs. https://intuitionlabs.ai/articles/ai-impact-graduate-jobs-2025
[34] “Canaries in the Coal Mine? Six Facts about the Recent Decline in Entry-Level Employment,” Stanford Digital Economy Lab (August 2025). https://digitaleconomy.stanford.edu/wp-content/uploads/2025/08/Canaries_BrynjolfssonChandarChen.pdf
[35] “Measuring US workers’ capacity to adapt to AI-driven job displacement,” Brookings Institution (2025). https://www.brookings.edu/articles/measuring-us-workers-capacity-to-adapt-to-ai-driven-job-displacement/
[36] “Reshaping work: Navigating the AI-driven labour market,” UNICEF Office of Research Innocenti, 2026 Global Outlook. https://www.unicef.org/innocenti/stories/2026-global-outlook-reshaping-work-ai-driven-labour-market