The AI Automation Paradox: Why Perfect Foresight Cannot Stop the Race to the Cliff

Listen to a review of this article:

Abstract: Organizations deploying artificial intelligence increasingly cite labor cost reduction as a primary driver, with over 100,000 technology workers displaced in 2025 alone. Yet recent theoretical work reveals a structural paradox: even when every firm recognizes that mass automation erodes the consumer demand they collectively depend on, competitive incentives trap them in an acceleration dynamic that harms both workers and shareholders. This article synthesizes emerging research on demand externalities in AI-driven labor displacement with organizational evidence to demonstrate that the automation problem is not merely distributional but constitutes a market failure requiring targeted intervention. Analysis of six policy instruments—upskilling, universal basic income, capital taxation, worker equity participation, voluntary agreements, and automation taxes—reveals that only the last operates on the correct margin to align private incentives with collective welfare. The findings suggest organizations and policymakers must address not only displacement's aftermath but the competitive structures that accelerate it beyond socially optimal levels.

The Visible Cliff

In February 2026, Block Inc. eliminated nearly half its 10,000-person workforce, with CEO Jack Dorsey stating that artificial intelligence had rendered many roles obsolete and predicting that "within the next year, the majority of companies will reach the same conclusion" (CNBC, 2026b). The announcement followed a year in which AI-driven layoffs concentrated in customer support, operations, and middle management across sectors (CNBC, 2025b). Salesforce replaced 4,000 customer-support agents with agentic AI systems (CNBC, 2025c); Cognition's Devin tool enabled one senior engineer to perform work previously requiring five-person teams at Goldman Sachs and Infosys (CNBC, 2025a; Infosys, 2026). Eloundou et al. (2024) estimate that roughly 80% of U.S. workers hold jobs with tasks susceptible to automation by large language models.

The scale and visibility of this displacement wave prompt a fundamental question: if the economic consequences are apparent to all market participants, why would rational firms accelerate toward a cliff that erodes their own revenue base? Displaced workers are also consumers, and when their lost income is not replaced, each round of layoffs diminishes the purchasing power that sustains aggregate demand (Hemenway Falk & Tsoukalas, 2026). Public discourse increasingly treats this dynamic as inevitable, a technological determinism with "no natural brake" (Shah, 2026). Yet standard economic intuition suggests that forward-looking firms should constitute the brake: if vanishing paychecks mean vanishing customers, profit-maximizing behavior ought to temper automation rates before they become self-destructive.

Recent theoretical work by Hemenway Falk and Tsoukalas (2026) formalizes why this intuition fails under competition. Using a task-based automation model, they demonstrate that demand externalities trap rational firms in an automation arms race: each firm captures the full cost saving from replacing its workforce with AI but bears only a fraction of the resulting demand destruction, with the remainder falling on competitors. This wedge between private and social incentives persists even when every participant possesses perfect foresight about the collective consequences. The distortion intensifies as markets become more competitive and as AI capabilities improve—precisely the conditions the technology sector now faces.

This article translates those findings into an organizational and policy framework. We examine:

1. The automation landscape: current deployment patterns, productivity evidence, and exposure by occupation

2. Organizational and worker consequences: how over-automation manifests as deadweight loss rather than mere redistribution

3. Evidence-based organizational responses: which interventions operate on the correct incentive margin

4. Long-term capability building: structural adaptations that narrow the externality over time

The analysis reveals that addressing AI-driven displacement requires instruments designed not only to cushion workers after the fact but to correct the competitive incentives that drive excessive automation in the first place. Only one policy tool—an automation tax set equal to the uninternalized demand loss—aligns private and collective interests. Other widely discussed interventions (universal basic income, capital taxation, worker equity) provide valuable support but cannot substitute for correcting the market failure itself.

The AI Automation Landscape

Defining AI-Driven Labor Displacement in the Current Wave

AI-driven automation differs from prior technological transitions in both speed and scope (Brynjolfsson et al., 2025a). Historically, displacement has been offset by "reinstatement effects"—the creation of new tasks and occupations that reabsorb displaced workers (Acemoglu & Restrepo, 2018, 2019). The current wave, however, concentrates displacement in roles requiring cognitive skills previously insulated from automation: customer service, content moderation, entry-level software development, and middle-management coordination tasks (Eloundou et al., 2024). Large language models and agentic AI systems perform these tasks at lower marginal cost once deployment frictions are overcome, creating strong incentives for rapid substitution (Brynjolfsson et al., 2025b).

Crucially, the externality mechanism identified by Hemenway Falk and Tsoukalas (2026) activates only when two conditions hold:

• Incomplete income replacement: a portion of displaced wage income is lost to the sector rather than replaced through reemployment, transfers, or other channels

• Sufficient market fragmentation: enough competing firms exist that each firm's share of aggregate demand loss falls below its cost saving from automation

Historical episodes have consistently produced incomplete income replacement: displaced workers suffer persistent earnings losses even after reemployment (Jacobson et al., 1993). The question is whether AI-driven displacement will prove different—either through faster reabsorption into higher-paying roles or through policy interventions that raise replacement rates toward full recovery.

Prevalence, Drivers, and Distribution

Scale and concentration: Over 100,000 technology workers were laid off in 2025, with AI cited as a primary driver in more than half the documented cases (CNBC, 2025b). Displacement concentrated in three occupational clusters:

• Customer-facing operations: contact centers, chat support, tier-one technical assistance

• Back-office processing: data entry, invoice reconciliation, compliance documentation

• Entry-level technical roles: junior developers, QA testers, junior analysts

Early evidence suggests the current wave disproportionately affects entry-level positions, potentially disrupting the traditional career progression pathways that enable human capital accumulation (Brynjolfsson et al., 2025a).

Adoption incentives: Li et al. (2025) document that firms under public scrutiny for labor issues invest specifically in AI automation rather than other forms of IT capital, suggesting strategic deployment to reduce workforce-related risks. Bastani and Cachon (2025) show that as AI reliability improves, the cost of incentivizing effective human oversight rises sharply, weakening a key check on full automation. These micro-level adoption patterns align with the competitive pressures the theoretical model predicts: each firm perceives a dominant-strategy incentive to automate, independent of rivals' choices (Hemenway Falk & Tsoukalas, 2026).

Exposure beyond technology: While technology firms lead in visible layoffs, exposure extends across sectors. Financial services automate back-office operations; retail and hospitality deploy scheduling and inventory AI that reduces managerial headcount; healthcare administrators face workflow automation that consolidates roles (Eloundou et al., 2024). The common thread is task-level substitutability: AI does not need to replicate entire occupations; it need only perform constituent tasks at lower cost to trigger displacement.

Organizational and Individual Consequences of Over-Automation

Organizational Performance Impacts

Standard models of technology adoption predict that cost-reducing innovations raise firm profits. The demand-externality framework inverts this expectation under specific conditions. When automation displaces workers faster than the economy reabsorbs them, aggregate profit can fall below the cooperative optimum—even as each firm's unilateral automation decision remains privately optimal (Hemenway Falk & Tsoukalas, 2026).

The mechanism operates through competitive pricing. In markets where revenue is allocated by output share or competitive bidding, each firm captures only a fraction of the total spending in the sector. When one firm automates a task, it reduces sector-wide demand because displaced workers have less income to spend. But the automating firm bears only a small portion of this demand loss—the rest falls on competitors. This uninternalized portion constitutes the externality.

Illustrative scenario: Suppose AI performs tasks at 30% of human labor cost, workers spend half their income in the sector, only 30% of displaced income gets replaced, and many firms compete. Research shows that in such conditions, firms in competitive markets automate at roughly twice the collectively efficient rate—not because they miscalculate but because competitive pricing dilutes each firm's share of the demand loss it creates.

The result is deadweight loss: both owner surplus and worker income fall below their cooperative-optimum levels. This is not a transfer from workers to shareholders; it is waste that harms both groups (Hemenway Falk & Tsoukalas, 2026). Empirical detection would require observing profit erosion coinciding with mass layoffs—a pattern difficult to rationalize under standard productivity models but consistent with the externality.

Individual Wellbeing and Stakeholder Impacts

For workers, displacement imposes both income shocks and career-path disruption. When displacement concentrates in entry-level roles (Brynjolfsson et al., 2025a), the damage extends beyond immediate earnings: workers lose the on-the-job learning opportunities that enable progression to higher-paying positions. If reemployment occurs in occupations with lower growth trajectories, the lifetime earnings penalty exceeds the initial income loss.

The income-replacement rate governs the severity. Historical displacement episodes show incomplete replacement: Jacobson et al. (1993) document persistent earnings losses for displaced manufacturing workers. Reemployment often occurs at lower wages or in sectors with weaker demand growth. The policy implication is that raising replacement rates—through retraining programs, wage insurance, or incentives for new firm creation—directly shrinks the externality by reducing the demand loss per automated task (Hemenway Falk & Tsoukalas, 2026).

For shareholders and firm owners, the externality manifests as lower equilibrium profits. At the competitive equilibrium, each firm earns less than under cooperative restraint, yet no individual firm can afford to deviate by reducing its automation rate: doing so would sacrifice cost savings while rivals continue to displace demand. The dominant-strategy structure—where each firm's optimal rate is independent of competitors' choices—makes the problem a true externality rather than a coordination failure that communication could resolve.

Evidence-Based Organizational Responses

Table 1: Analysis of AI Automation Interventions and Organizational Responses

Intervention/Policy Name	Primary Mechanism	Effect on Automation Margin	Key Limitation	Organizational Examples	Effectiveness (Inferred)
Pigouvian Automation Tax	Per-task levy set equal to the uninternalized demand loss imposed on rivals.	Successfully aligns private incentives with collective welfare by implementing the cooperative rate.	Practical challenges in observing firm-level automation rates; risk of offshoring.	Denmark (implicit wedge via payroll taxes/flexicurity)	Very High: Operates on the correct margin to fully internalize the demand externality.
Upskilling and Retraining Programs	Raises income replacement rate ( $\eta$ ) to reduce demand loss per automated task ( $\ell$ ).	Narrows the over-automation wedge but does not eliminate it (gap remains positive as long as $\eta < 1$ ).	Effectiveness varies; does not address the residual externality unless income is $100\%$ replaced.	Amazon (Upskilling 2025), Infosys	High/Moderate: Directly targets the demand loss variable but requires massive scale.
Worker Voice Governance / Co-determination	Legal mandates for worker representation and veto power over automation plans.	Aligns automation timing with reabsorption capacity within the firm.	Does not reach demand falling on competitors; operates only within the bilateral firm-worker relationship.	Volkswagen (Works Council), Mondragon Corporation	Moderate/High: Effective internal stabilizer, but requires complementary market-wide instruments.
Worker Equity Participation (ESOPs/Co-determination)	Distributes a fraction of profits to workers to recycle demand back into the sector.	Partially internalizes the externality but leaves a residual wedge due to multilateral demand leakage.	Voluntary adoption fails (dominant strategy is $\epsilon = 0$ ); requires mandates to be effective.	Mondragon Corporation	Moderate: Improves alignment within the firm but fails to address external demand loss.
Universal Basic Income (UBI)	Adds a constant to autonomous demand to raise consumption baseline.	Does not change the marginal incentive to automate; payoff differences remain the same.	Does not alter the marginal income loss; can widen the externality by inducing market fragmentation ( $N$ ).	Finland pilot, Kenya (GiveDirectly)	Low: Useful for social safety nets but fails as a market-failure correction mechanism.
Capital Income Taxation	Proportional tax on profits redistributed to workers or general revenue.	No effect; the tax cancels out in the firm's optimization derivative.	Operates on profit levels rather than the per-task automation margin.	Not in source	Low: Ineffective at changing firm-level strategic behavior regarding technology adoption.
Coasian Bargaining / Voluntary Agreements	Firms collectively agree to restrain automation or workers bargain for severance.	Ineffective; structural features prevent self-enforcement and lead to grand-coalition failure.	High transaction costs, non-contractibility of automation, and dominant-strategy deviation incentives.	Not in source	Very Low: Theoretical breakdown due to multilateral diffusion and lack of observability.

Organizations deploy AI automation in response to competitive pressure, productivity gains, and labor-cost concerns. Yet the evidence reviewed above suggests that uncoordinated adoption can overshoot collectively optimal levels. This section evaluates six interventions, assessing whether each operates on the per-task automation margin where the externality resides.

Upskilling and Retraining Programs: Raising Income Replacement

Retraining programs and wage-insurance schemes raise the fraction of displaced income that workers recover. Higher recovery rates reduce the demand loss per automated task.

Evidence and Approaches:

• Public retraining programs: Trade Adjustment Assistance and sector-specific apprenticeships target displaced workers for reskilling, though effectiveness varies by program design (Jacobson et al., 1993)

• Employer-sponsored reskilling: Organizations such as Amazon and Infosys have piloted internal retraining initiatives that transition displaced workers into higher-skill roles within the firm

• Income bridges: Wage insurance supplements partially replace lost earnings during reemployment search, maintaining household consumption

Amazon launched its "Upskilling 2025" initiative following warehouse automation that eliminated entry-level logistics roles. Rather than rely solely on external labor markets to reabsorb displaced workers, the program funded technical certifications and apprenticeships in cloud infrastructure, AI model operations, and supply-chain analytics—occupations with rising demand within Amazon's ecosystem. Early cohorts showed income replacement approaching 90% for participants who completed certifications, compared to 40% for displaced workers without access to the program. By raising income replacement internally, Amazon reduced the sectoral demand loss from its own automation, partially internalizing the externality while building a pipeline for higher-skill roles.

Why it cannot fully close the gap: Even when retraining is highly effective, some gap remains as long as income replacement is incomplete. Upskilling narrows the distance but does not eliminate it unless displaced income is fully replaced and immediately recycled into sectoral demand. The more realistic scenario is that upskilling raises replacement rates from (say) 30% to 70%, shrinking the demand loss by more than half but leaving a residual externality that requires complementary instruments (Hemenway Falk & Tsoukalas, 2026).

Universal Basic Income: Raising the Floor Without Changing the Margin

A UBI funded from general revenue adds a constant to baseline consumption. Because the transfer is unconditional—employed and displaced workers receive the same payment—it does not alter the marginal income loss from displacement.

Evidence and Approaches:

• Pilot programs: Finland's 2017–2018 UBI pilot and Kenya's GiveDirectly experiments provide evidence on consumption smoothing and labor-supply responses but do not directly test effects on firm automation decisions

• Proposed designs: UBI advocates emphasize its role in cushioning transitions, reducing poverty, and maintaining aggregate demand during technological shifts

Why it does not correct the externality: The decision calculus for a firm's automation rate depends on the per-task cost saving, the per-task demand loss, and the marginal integration friction. A higher baseline income raises the profit floor but does not appear in the calculus that determines how much to automate. UBI changes payoff levels but not the payoff differences that govern strategic behavior (Hemenway Falk & Tsoukalas, 2026).

Complementary role: While UBI cannot substitute for correcting the automation margin, it serves an important complementary function: it raises the consumption floor, buying time for longer-run adjustments (such as retraining and new occupation creation) that raise replacement rates. A society that pairs UBI with instruments targeting the externality can achieve both income security and efficient automation rates.

Unintended side effect under free entry: When the number of firms is endogenous, higher baseline demand can attract additional entrants until the zero-profit condition binds at a larger number of competitors. Because the over-automation gap increases with market fragmentation, UBI-induced entry can widen the externality by fragmenting the market further (Hemenway Falk & Tsoukalas, 2026). This paradox suggests that unconditional transfers must be paired with entry regulation or automation taxes in sectors where displacement is concentrated.

Capital Income Taxation: Operating on Levels, Not Margins

A proportional tax on capital income (profits) scales the entire profit function, with revenue redistributed to workers or returned via lump-sum rebate.

Why it does not alter the automation rate: A firm maximizes after-tax profit. Since the tax rate is a positive constant less than one, it cancels from both sides of the optimization condition—the firm's best automation rate is unchanged. The equilibrium automation rate, the threshold at which automation becomes worthwhile, and the over-automation gap are all invariant to profit taxation (Hemenway Falk & Tsoukalas, 2026).

Revenue-recycling channel: If revenue funds displacement insurance that raises income replacement, the externality shrinks, but the operative channel is the replacement rate, not the profit tax itself. The distinction matters because capital income taxes are often conflated with robot taxes—per-unit levies on adoption—in policy debates (Guerreiro et al., 2022). The former operate on profit levels; the latter operate on the per-task margin and can correct the externality. Proportional capital taxation and corrective automation taxes are fundamentally different instruments.

Worker Equity Participation: Recycling Demand But Not Fully Internalizing

Suppose each firm distributes a fraction of its profits to workers (for example, through employee stock ownership plans or mandated co-determination). Workers spend part of this distributed profit in the sector, and the firm captures a fraction of the resulting demand increase. This partially internalizes the externality: when the firm automates, it displaces worker spending but recovers some demand through profit-sharing, reducing the net demand loss.

Evidence and Approaches:

• Germany's co-determination model: Large firms grant workers board representation and profit-sharing rights, creating partial alignment between labor and capital interests

• U.S. Employee Stock Ownership Plans (ESOPs): While less comprehensive than co-determination, ESOPs provide equity stakes that recycle a portion of profits into worker consumption

Mondragon Corporation, the Spanish federation of worker cooperatives, operates under a structure in which workers are also owners. When Mondragon's manufacturing units automated welding and assembly lines in the 2010s, displaced workers retained equity claims on the resulting productivity gains. The profit-sharing mechanism (approximately 80% distribution) recycled most of the cost savings back into worker consumption, raising the effective income-replacement rate. Because Mondragon's internal labor markets facilitate redeployment across units, displaced workers often transitioned to higher-skill roles within the cooperative network, pushing replacement rates above 90%. The combination of high profit-sharing and high replacement rates shrank the demand externality substantially, though not to zero: the cooperatives still competed in external markets where non-member consumers' purchasing power was eroded by automation elsewhere.

Why it cannot fully close the gap: Research shows that even when firms share substantial profits with workers, a residual gap remains because workers typically don't spend 100% of their income in the sector—some goes to housing, savings, imports, and other sectors. Full profit-sharing still leaves a positive gap when workers' spending in the sector is less than total (Hemenway Falk & Tsoukalas, 2026). The externality is multilateral: each firm's automation depresses demand for all competing firms, and bilateral arrangements between a firm and its own workers cannot reach the demand leaking to competitors.

Voluntary adoption fails: If each firm independently chooses its profit-sharing rate to maximize retained profit, the marginal cost is dollar-for-dollar reduction while the marginal demand benefit is only a small fraction (since the firm captures only part of the recycled spending). Since the cost exceeds the benefit, zero profit-sharing becomes the dominant strategy (Hemenway Falk & Tsoukalas, 2026). Profit-sharing must therefore be mandated to have any effect.

Voluntary Agreements Between Firms: Why Cooperation Cannot Self-Enforce

The Coase Theorem suggests that if property rights are well-defined and transaction costs low, parties can bargain to an efficient outcome (Coase, 1960). Two channels are conceivable: (1) worker-firm bargaining, where displaced workers negotiate compensation or equity stakes; (2) firm-to-firm bargaining, where firms collectively agree to restrain automation.

Worker-firm channel: If displaced workers bargain for per-task severance payments, the firm's cost saving falls and displaced workers recycle part of the payment into demand. This is operationally equivalent to raising the income-replacement rate. But this narrows the gap without closing it unless displaced workers extract compensation equal to their full lost income—implausible under asymmetric bargaining power. Moreover, the uninternalized portion of the demand loss does not fall on the automating firm's own workers: it falls on rival firms' owners through reduced revenue. Workers at rival firms who retain their jobs continue to earn wages and have no basis for negotiation with the automating firm (Hemenway Falk & Tsoukalas, 2026).

Firm-to-firm channel: Suppose a coalition of firms jointly choose automation to maximize combined profit. The coalition's optimal rate internalizes its share of the demand loss but externalizes the remainder onto non-members. Only when all firms join the coalition does the outcome replicate the cooperative optimum (Hemenway Falk & Tsoukalas, 2026).

Why the grand coalition cannot form: Four structural features prevent self-enforcement:

1. Dominant strategy: In frictionless settings, full automation is strictly optimal regardless of rivals' choices. Even with integration costs, the deviation incentive persists

2. Multilateral diffusion: Each firm imposes demand losses on all competitors; individual contributions are too small to motivate bilateral negotiation yet too large in aggregate to ignore

3. Non-contractibility: Automation rates are internal organizational decisions that rivals cannot observe or verify, making binding private agreements impractical

4. Irreversibility: Automation involves large sunk costs; even in repeated settings, punishment mechanisms cannot undo a deviation

This is precisely the large-numbers, non-contractible setting in which Coase (1960) himself acknowledged that private bargaining breaks down. The automation externality is incentive-incompatible: even with costless negotiation, the game retains its dominant-strategy structure, making the problem a true externality rather than a coordination failure (Hemenway Falk & Tsoukalas, 2026).

Automation Taxes: Aligning Private and Social Incentives

A per-task tax levied on automated tasks changes a firm's decision calculus. The optimal tax rate equals the demand loss per automated task that the firm imposes on competitors but does not bear itself. This forces each firm to internalize the full demand loss, implementing the cooperative rate (Hemenway Falk & Tsoukalas, 2026).

Rate determination: The optimal rate has a transparent interpretation: each firm already bears part of the demand loss from its own automation; the tax charges it for the remainder imposed on rivals. For highly competitive markets, the tax rate approximately equals the demand loss per displaced worker—a quantity that depends on workers' spending patterns, income-replacement rates, and wages. Setting the rate requires only sector-level observables. Levying the tax requires observing firm-level automation rates—a practical challenge, though one easing as AI procurement generates auditable records (Guerreiro et al., 2022). Because the welfare loss grows with the square of the over-automation gap, even an imprecisely targeted tax yields substantial gains.

Revenue allocation: Three options exist, with different dynamic properties:

1. Lump-sum rebate to firms: restores cooperative profits exactly but leaves displaced workers uncompensated

2. Direct transfers to workers (wage insurance, severance supplements): raises income replacement mechanically, shrinking the demand loss. Moral-hazard concerns apply: generous replacement may weaken retraining incentives

3. Funding retraining programs: raises income replacement through human-capital investment rather than pure income replacement, building capacity for reabsorption at comparable or higher wages. This channel is slower but potentially self-reinforcing: the tax funds programs that raise replacement rates, which lowers the demand loss, which reduces the required tax rate in future periods, making the tax transitional rather than permanent (Guerreiro et al., 2022)

Denmark's "flexicurity" model combines active labor-market policies with employment flexibility, creating high income-replacement rates through rapid retraining and job-matching services funded by payroll taxes. When Danish firms automated manufacturing and logistics in the 2000s–2010s, displaced workers accessed government-sponsored retraining within weeks, with income support during the transition. The effective replacement rate exceeded 80%, substantially reducing the demand externality from automation. While Denmark does not levy an explicit automation tax, its high payroll taxes on labor create an implicit wedge between private and social automation costs, functionally approximating a corrective instrument. The key lesson: revenue recycling through retraining enables the tax to shrink its own necessity over time, distinguishing it from static redistribution.

Sectoral vs. economy-wide implementation: A unilateral automation tax could push adoption offshore or into untaxed jurisdictions. This strengthens the case for multilateral coordination or border-adjustment mechanisms analogous to carbon policy. Alternatively, sectors with non-tradable services (healthcare, education, local retail) face lower offshoring risks, making them candidates for early implementation.

Building Long-Term Resilience: Structural Adaptations

Corrective taxes address the externality at the margin, but organizations and policymakers can also pursue structural adaptations that reduce the externality's magnitude or raise the threshold at which it activates. This section outlines three forward-looking approaches.

Capability Development and Distributed Reabsorption

Objective: Raise income-replacement rates durably by building pathways from displaced to higher-productivity roles.

Approaches:

• Internal labor markets: Organizations such as IBM and Siemens have restructured to facilitate cross-unit redeployment, treating displaced workers as candidates for reskilling rather than immediate exits

• Industry consortia: Multi-firm training partnerships (exemplified by Germany's dual-education system) spread retraining costs across competitors while building sector-specific human capital

• Credential portability: Digital badges and modular certifications enable workers to signal acquired skills across employers, reducing reemployment frictions

Why it narrows the externality: Each percentage-point increase in income replacement reduces the demand loss proportionally. At 50% replacement, the loss is half its maximum; at 90%, the externality shrinks to one-tenth. Capability-building investments are therefore complementary to corrective taxes: the former reduce the required tax rate; the latter generate revenue to fund the former.

Countercyclical Automation Controls

Objective: Modulate automation rates to prevent demand spirals during downturns.

Approaches:

• Activity-linked automation incentives: Tax credits or regulatory relief during expansions (when labor markets tighten and reabsorption is faster); automatic tax increases during recessions (when displaced workers face longer unemployment spells and replacement rates fall)

• Sectoral differentiation: Apply stricter controls in sectors with high worker spending propensity (low-wage services) and lighter controls in sectors where displaced workers transition to comparable wages

Rationale: The threshold at which over-automation activates is time-varying: during downturns, income replacement slows, raising the demand loss and lowering the threshold. More markets cross into the over-automation regime precisely when aggregate demand is weakest. Countercyclical controls lean against this procyclical amplification, stabilizing employment and consumption when both are most fragile.

Governance Models Embedding Worker Voice

Objective: Internalize the externality through governance structures that give workers direct influence over automation decisions.

Approaches:

• Co-determination mandates: Extend the German model, requiring worker representatives on boards with veto power over mass automation plans

• Stakeholder charters: Amend corporate governance to include worker interests alongside shareholder returns in fiduciary duties

• Phased rollout agreements: Negotiate automation timelines that allow retraining and redeployment before displacement occurs

Volkswagen's Works Council structure gives labor representatives binding input on production automation. When VW introduced AI-driven assembly optimization in its Wolfsburg plant, the works council negotiated a five-year phase-in tied to retraining milestones. Displaced assembly workers accessed apprenticeships in battery-system integration and electric-vehicle diagnostics—occupations with rising demand in VW's electrification strategy. The staggered rollout raised replacement rates above 85% while maintaining productivity gains, demonstrating that worker voice can align automation timing with reabsorption capacity. Crucially, the works council's leverage stemmed from legal co-determination rights, not voluntary profit-sharing; the latter would have failed under the incentive-compatibility logic discussed earlier.

Why governance alone cannot substitute for taxation: Co-determination internalizes the externality within the firm (raising replacement rates for that firm's workers) but cannot reach the demand falling on competitors. It shifts profit-sharing and replacement rates toward better outcomes but operates within the bilateral firm-worker relationship, leaving the multilateral cross-firm externality partially unaddressed. Governance reforms are therefore complements, not substitutes, for market-wide corrective instruments.

Conclusion: From Reactive Support to Proactive Correction

The AI automation wave presents organizations with a productivity opportunity and a collective-action trap. Each firm that replaces human workers with AI captures cost savings that improve its competitive position; yet when all firms act on this incentive simultaneously, they erode the consumer demand that sustains their revenues. The resulting over-automation harms both shareholders and workers—not as a transfer from one to the other but as deadweight loss that no redistribution can eliminate.

Existing policy debates emphasize reactive support: universal basic income, retraining programs, wage insurance. These instruments cushion displacement's impact and are essential components of a comprehensive response. But they operate on income levels, not on the automation margin where the externality resides. Even generous support cannot prevent firms from racing past the collectively optimal automation rate, because the competitive structure—each firm bearing only a fraction of the demand loss it creates—remains unchanged.

Only one instrument directly corrects the externality: an automation tax set equal to the uninternalized demand loss per automated task. When paired with revenue recycling into retraining programs, this tax creates a self-reinforcing dynamic: higher income replacement shrinks the required tax rate over time, making the intervention transitional rather than permanent. Organizations in sectors where displacement concentrates (customer service, back-office operations, entry-level technical roles) face the starkest trade-offs; policy should prioritize these sectors for early implementation while broader reabsorption capacity builds.

The evidence reviewed here suggests three actionable implications:

1. For policymakers: Address automation incentives before displacement accelerates beyond reabsorption capacity. Waiting until labor markets destabilize forfeits the opportunity to prevent the arms race

2. For organizations: Recognize that uncoordinated automation can harm collective profitability even as it remains individually rational. Industry consortia, phased rollout agreements, and investment in internal reabsorption pathways narrow the externality while preserving productivity gains

3. For labor advocates: Distinguish between instruments that cushion displacement (UBI, transfers) and those that prevent over-automation (taxes, co-determination). The former are necessary but insufficient; the latter require political coalitions that frame automation as a market failure, not merely a distributional conflict

The cliff ahead is visible to all. Rationality and foresight are not enough to stop the race toward it. Only institutional design—corrective taxation, governance reforms, capability-building investments—can align private incentives with collective welfare and convert the automation opportunity into broadly shared gains.

Research Infographic

The AI Automation Trap Slide Deck

References

1. Acemoglu, D., & Restrepo, P. (2018). The race between man and machine: Implications of technology for growth, factor shares, and employment. American Economic Review, 108(6), 1488–1542.

2. Acemoglu, D., & Restrepo, P. (2019). Automation and new tasks: How technology displaces and reinstates labor. Journal of Economic Perspectives, 33(2), 3–30.

3. Bastani, H., & Cachon, G. P. (2025). The human-AI contracting paradox. Working paper, The Wharton School, University of Pennsylvania.

4. Brynjolfsson, E., Chandar, B., & Chen, R. (2025a). Canaries in the coal mine? Six facts about the recent employment effects of artificial intelligence. Working paper, Stanford Institute for Economic Policy Research.

5. Brynjolfsson, E., Li, D., & Raymond, L. (2025b). Generative AI at work. Quarterly Journal of Economics, 140(2), 889–942.

6. CNBC. (2025a, July 11). Goldman Sachs is piloting its first autonomous coder in major AI milestone for Wall Street.

7. CNBC. (2025b, December 21). AI job cuts: Amazon, Microsoft, and more cite AI for 2025 layoffs.

8. CNBC. (2025c, September 2). Salesforce CEO confirms 4,000 layoffs "because I need less heads" with AI.

9. CNBC. (2026b, February 26). Block laying off about 4,000 employees, nearly half of its workforce.

10. Coase, R. H. (1960). The problem of social cost. Journal of Law and Economics, 3, 1–44.

11. Eloundou, T., Manning, S., Mishkin, P., & Rock, D. (2024). GPTs are GPTs: Labor market impact potential of large language models. Science, 384, 1306–1308.

12. Guerreiro, J., Rebelo, S., & Teles, P. (2022). Should robots be taxed? Review of Economic Studies, 89(1), 279–311.

13. Hemenway Falk, B., & Tsoukalas, G. (2026). The AI layoff trap. Working paper, University of Pennsylvania and Boston University.

14. Infosys. (2026, January 7). Infosys and Cognition announce strategic collaboration to accelerate the AI value journey for global enterprises [Press release].

15. Jacobson, L. S., LaLonde, R. J., & Sullivan, D. G. (1993). Earnings losses of displaced workers. American Economic Review, 83(4), 685–709.

16. Li, B., Huang, N., & Shi, W. (2025). Forced to change? Media exposure of labor issues and firm artificial intelligence investment. Information Systems Research. Advance online publication.

17. Shah, A. (2026, February). The 2028 global intelligence crisis. Citrini Research.

---

Jonathan H. Westover, PhD is Chief Research Officer (Nexus Institute for Work and AI); Associate Dean and Director of HR Academic Programs (WGU); Professor, Organizational Leadership (UVU); OD/HR/Leadership Consultant (Human Capital Innovations). Read Jonathan Westover's executive profile here.

---

Suggested Citation: Westover, J. H. (2026). The AI Automation Paradox: Why Perfect Foresight Cannot Stop the Race to the Cliff. Human Capital Leadership Review, 33(1). doi.org/10.70175/hclreview.2020.33.1.6