Every claim. Every source. Every limitation acknowledged.

The decomposition of post-war productivity growth into capital deepening, total factor productivity, and labour quality contributions is uncontroversial in macroeconomics. Standard references: the OECD Productivity Statistics database, the U.S. Bureau of Labor Statistics Multifactor Productivity series, and the Penn World Tables.

Across OECD countries between 1970 and 2005, capital deepening contributed 0.8 to 1.2 percentage points annually to labour productivity growth. Total factor productivity contributed an additional 0.7 to 1.0 points. Labour quality improvements contributed 0.2 to 0.4 points.

The U.S. Bureau of Labor Statistics has documented that between 1973 and 2014, labour productivity in the nonfarm business sector grew at 1.7 percent per year, with capital deepening accounting for approximately 60 percent of that growth.

Workers were the platform on which the productivity gains were deployed. The same human, with better machines and better software, produced more output per hour. The cognitive demand on the worker was approximately constant; what changed was the leverage their hour created.

The exact percentages of the decomposition vary by methodology, country, and time period. The "75 years" framing anchors to the post-WWII baseline (roughly 1948 onwards) and reflects the period over which the capital-and-technology productivity model held; the highest gains were concentrated in the early decades, with the slowdown beginning around 2005. The Solow growth model that underlies these decompositions has known limitations, particularly around treatment of human capital and intangible investment.

• OECD Productivity Statistics. oecd.org/sdd/productivity-stats
• U.S. BLS Multifactor Productivity. bls.gov/productivity
• Penn World Tables. rug.nl/ggdc/productivity/pwt

The U.S. Bureau of Labor Statistics has documented the productivity slowdown extensively. The post-2004 productivity growth rate of 1.4 percent per year (through 2014) was 30 percent below the 1948 to 1973 average. The most recent BLS analyses (2023 to 2025) have continued to show weak capital deepening contributions to overall productivity.

The OECD has published multiple studies documenting that across nearly every OECD country, productivity growth slowed sharply after 2005. The cumulative cost of the U.S. productivity slowdown from 2005 to 2018 has been estimated at $10.9 trillion in lost output.

• BLS productivity slowdown analysis. bls.gov/opub/mlr/2018
• OECD Compendium of Productivity Indicators (annual).
• The Productivity Slowdown. Federal Reserve Bank of San Francisco conference, 2017 and subsequent updates.

Microsoft Research, January 2025. "The Impact of Generative AI on Critical Thinking" (Lee et al., 2025), based on a survey of 319 knowledge workers, found that AI use shifts cognitive work from generation to evaluation. Heavy AI users showed reduced critical thinking engagement during routine tasks, but the cognitive work for non-routine tasks was reported as more demanding when AI was involved. Workers expended more effort applying critical thinking with AI than they would performing the same tasks without it.

Harvard Business Review, February 2026. Eight months of research inside a 200-person U.S. tech firm found that AI tool adoption was correlated with increased work intensity rather than decreased workload. Employees using AI tools did not work less; output expectations rose to consume the time savings.

UC Berkeley Labor Center, 2025. A longitudinal study tracked workers who adopted AI tools through 2025. 67 percent of adopters reported working more hours, not fewer, by the end of the year. This is the single most cited finding in the AI cognitive load literature because it is longitudinal and specifically isolates the adopter population.

Deloitte, 2025 Workforce Intelligence Report. Mental fatigue and cognitive strain have surpassed workload volume as the leading predictors of burnout. This is a methodological shift; previously workload volume was the primary predictor, and the change indicates that the nature of work has shifted toward higher cognitive demand independent of hours worked.

Three structural mechanisms explain why AI increases cognitive load rather than decreasing it.

Elimination of cognitive recovery breaks. Pre-AI knowledge work contained natural cognitive breaks: waiting for reports to compile, manually formatting documents, searching through documents for specific data points. These tasks were not intellectually demanding and served as built-in recovery periods. AI eliminates these breaks. When every task that used to take twenty minutes now takes twenty seconds, the worker moves immediately to the next cognitively demanding task. The result is an uninterrupted stream of high-intensity cognitive work with no natural recovery time.

Multiplication of decisions. AI multiplies decisions rather than eliminating them. Every output requires an accept, reject, or revise decision under uncertainty. Pre-AI, decisions were embedded in the writing process. Post-AI, the worker generates outputs faster but must still evaluate, edit, validate, and approve each one. The cognitive load shifts from generation to evaluation, which is often more draining for sustained periods.

Rising expectation curve. Marketing departments using AI content generation have been expected to produce 3.2x more content pieces per month compared to pre-AI baselines (Content Marketing Institute, 2025). As AI makes individual tasks faster, the volume of expected output rises to consume the time savings. The worker is now responsible for managing 3x the output with no proportional increase in cognitive bandwidth.

• Lee, M., et al. (2025). The Impact of Generative AI on Critical Thinking. Microsoft Research. microsoft.com/research
• Harvard Business Review (2026). 200-person tech firm AI adoption study.
• UC Berkeley Labor Center (2025). AI Adoption Longitudinal Study.
• Deloitte (2025). Workforce Intelligence Report.
• Content Marketing Institute (2025). AI Content Production Benchmarks.

Gallup, State of the Global Workplace 2026. Global engagement is at 20 percent, the lowest measured level since 2020. Manager engagement dropped 9 percentage points between 2022 and 2025, with a 5-point drop between 2024 and 2025 alone. The historical "manager engagement premium" between manager engagement and individual contributor engagement has effectively closed.

ActivTrak, State of the Workplace 2026. Behavioral data from 163,638 employees across 1,111 companies. Disengagement risk now exceeds burnout risk for the first time on record. Focus efficiency dropped to 60 percent.

Perceptyx longitudinal motivation data. Drawing on 20 million employee survey responses, Perceptyx has documented intrinsic motivation collapse over a multi-year window.

• Gallup. State of the Global Workplace 2026. gallup.com
• ActivTrak. State of the Workplace 2026.
• Perceptyx workforce research archives.

If capital deepening has slowed (worker hours less leveraged by capital), AI is increasing cognitive demand (worker hours more cognitively expensive), and discretionary effort has collapsed (workers withholding the extra effort that used to absorb new demands), then the productivity model that worked from 1948 to 2005 has structurally broken. The three sub-claims combine into a coherent argument about why productivity now depends on the human in a way it did not for 75 years.

The three sub-claims are individually well-supported, but the integration into a single thesis is our framing rather than a published academic finding. Other framings of the same data are possible.

The capital deepening slowdown is empirically established; the cause is contested. The AI cognitive load research is recent (2025 to 2026) and may be subject to revision as longer-term studies emerge. The discretionary effort collapse is well-documented but the causal relationship to AI is correlational, not proven. We treat the integration as the strongest available framing of the available evidence rather than as a proven causal model.

This is the inferential move in the thesis. It is logically derived from Move 2 rather than directly cited from research.

If capital can no longer multiply hours (Move 2.2a), AI is consuming rather than freeing cognitive bandwidth (Move 2.2b), and discretionary effort is no longer quietly available (Move 2.2c), then the only remaining lever for productivity gains is human capacity itself.

This is a logical claim, not an empirical one. A skeptical reader can challenge the framing without needing to dispute the underlying data.

The claim is defensible because the alternatives have been structurally weakened. Capital deepening has slowed in nearly every OECD country. AI investment has not produced the expected productivity gains; Gallup's Q1 2026 data shows 89 percent of executives report no productivity impact from AI in the past three years. If the traditional productivity multipliers are not working, the human capacity is the remaining variable.

The phrase "for the first time in three generations" anchors to the 75-year span over which the post-WWII productivity model has held, with the highest gains concentrated in the early decades. The conversational shorthand is "three generations"; the underlying empirical span is 1948 to roughly 2023.

This is the move most vulnerable to the "AI will eventually deliver productivity gains" objection. The current data shows AI has not delivered, but five years of underdelivery is not the same as permanent failure. The thesis describes the present moment, not the long-term trajectory.

The depletion claim is supported by the engagement collapse data (Gallup 2026), the burnout data (Eagle Hill 2025: 71 percent middle manager burnout), the false-retention data (MetLife 2026: 56 percent staying out of necessity), and the disengagement-exceeds-burnout finding (ActivTrak 2026). The cohort data is Clover ERA's unique contribution, documented in detail in section 04 below.

The combination of higher cognitive demand (Move 2.2b) and reduced discretionary effort (Move 2.2c) produces depletion that is observable at the company level through perception gap measurement. Managers self-report higher capability than their teams report experiencing; that gap is the proxy measurement for depletion the dashboards don't catch.

The cohort sample size is small (n=11 as of Q1 2026, growing toward Q2 publication). Single-company variance is high; the median $20M is more reliable than any individual company estimate. The cohort represents companies in the 300 to 1,200 employee range across four sectors; findings may not generalize to smaller or larger companies, or to sectors not represented in the cohort.

The Manager Gap Index is a structured assessment that produces a score and a cost estimate based on documented research multipliers. It does not predict individual outcomes for specific people on specific teams.

The cost calculation is an estimate. The six cost layers use research-backed multipliers, but each multiplier has its own confidence interval. Companies should treat the cost number as directionally correct rather than as accounting precision.

The cohort represents companies in the 300 to 1,200 employee range across four sectors; findings may not generalize beyond this range. Sample sizes below n=3 in any segment cut are not reported.

A high MGI score correlates with the patterns described in the cohort findings, but the causal pathway from manager perception gap to financial outcome is correlational rather than proven.