The Competitive Trap: How AI-Driven Automation Creates Collective Market Failure

Listen to a review of this article:

Abstract: Recent evidence suggests that artificial intelligence is displacing workers at an accelerating pace across multiple industries, with over 100,000 technology workers laid off in 2025 alone due to AI adoption. This article examines a critical yet underappreciated market failure: when firms automate in competitive environments, each captures the full cost savings while bearing only a fraction of the resulting demand destruction, creating a strategic externality that harms both workers and firm owners. Drawing on game-theoretic models and recent empirical observations, we demonstrate that competitive pressure traps rational, forward-looking firms in an automation arms race that exceeds collectively optimal levels. Neither wage flexibility, profit-sharing arrangements, nor voluntary coordination mechanisms can eliminate this distortion. Only policy interventions that directly address the per-task automation margin—specifically, Pigouvian automation taxes calibrated to uninternalized demand losses—can restore efficiency. The analysis reveals that "better" AI paradoxically amplifies rather than resolves the problem, and that fragmented markets suffer disproportionately. These findings suggest policy discourse should shift from managing displacement consequences to correcting the competitive incentives driving excessive automation.

In February 2026, Block Inc. eliminated nearly half its workforce—approximately 5,000 positions—with CEO Jack Dorsey attributing the decision directly to AI capabilities that rendered those roles obsolete (CNBC, 2026b). The announcement included a stark prediction: "within the next year, the majority of companies will reach the same conclusion." This was not isolated hyperbole. Salesforce had already replaced 4,000 customer-support agents with agentic AI systems (CNBC, 2025c), while Cognition's Devin platform, deployed at Goldman Sachs and Infosys, enables one senior engineer to perform work previously requiring five-person teams (CNBC, 2025a; Infosys, 2026). Across 2025, over 100,000 tech workers lost positions to AI-driven automation, with the technology cited as the primary driver in more than half the cases (CNBC, 2025b).

The scale and velocity of this transformation have reignited longstanding debates about technological displacement. Since Ricardo's (1821) observations on mechanization and Keynes's (1930) musings on technological unemployment, economists have generally emphasized self-correcting mechanisms—particularly the "reinstatement effect" through which automation of existing tasks creates demand for new occupations (Acemoglu & Restrepo, 2018, 2019). Yet Autor et al. (2024) document that displacement has intensified over recent decades while job creation has not consistently kept pace, and early evidence suggests current AI adoption disproportionately affects entry-level positions that historically served as pathways to career advancement (Brynjolfsson et al., 2025a).

A fundamental puzzle emerges: if automation erodes the purchasing power that sustains demand for all firms' output, why do rational, forward-looking companies not restrain their adoption accordingly? Displaced workers are simultaneously consumers; when their income vanishes without replacement, aggregate spending contracts. At the extreme, this dynamic becomes self-destructive: firms automate their way to limitless productivity but zero customers. Public discourse increasingly frames this trajectory as inevitable (Shah, 2026), yet economic theory suggests firms should internalize these consequences and self-correct before reaching the precipice.

Recent theoretical work by Hemenway Falk and Tsoukalas (2026) demonstrates that this intuition fails systematically in competitive markets. Through a task-based model in which firms choose automation rates while facing demand feedback from worker displacement, they show that competition creates a structural externality: each automating firm captures full cost savings but bears only a fractional share of the resulting aggregate demand loss, with the remainder falling on competitors. Even with perfect foresight and common knowledge of the demand cliff ahead, profit-maximizing firms over-automate relative to what a cooperative agreement would dictate. The resulting surplus destruction harms both workers and firm owners—a deadweight loss rather than a mere transfer between factor classes.

This article translates those theoretical insights into actionable guidance for practitioners and policymakers. We examine how organizational responses to AI adoption interact with the underlying competitive dynamics, assess which policy instruments can correct the distortion, and identify structural features that amplify or mitigate the problem. The analysis reveals that many widely discussed interventions—universal basic income, capital income taxation, worker equity schemes, and voluntary bargaining—operate on the wrong margins and cannot eliminate the externality. Only interventions targeting the per-task automation decision itself, particularly Pigouvian taxes set equal to uninternalized demand losses, directly address the source of market failure.

The AI Automation Landscape

Defining AI-Driven Displacement in Competitive Markets

AI-driven displacement differs from previous automation waves in both scope and substitution capability. Where earlier technologies typically automated routine manual or cognitive tasks within narrow domains, large language models and agentic AI systems demonstrate competence across diverse knowledge work (Eloundou et al., 2024). Estimates suggest approximately 80% of U.S. workers hold positions with tasks susceptible to automation by current-generation models—a penetration rate far exceeding prior technological transitions (Eloundou et al., 2024).

The competitive dimension is equally critical. Firms do not adopt AI in isolation; they operate within industries where rivals face identical incentive structures. Li et al. (2025) document that firms experiencing labor-relations scrutiny specifically accelerate AI investments relative to other information technology spending, suggesting automation serves strategic competitive purposes beyond pure efficiency gains. Brynjolfsson et al. (2025b) confirm substantial productivity improvements from generative AI adoption in customer service contexts, establishing that private returns to automation are real and significant. Yet these micro-level productivity gains, observed at individual firms, do not account for system-level effects when all competitors simultaneously displace workers who constitute each other's customer base.

The task-based framework provides analytical clarity (Acemoglu & Restrepo, 2018). Within any firm, production requires performing numerous tasks—customer interaction, data processing, content generation, quality assurance, and so forth. Each task can potentially be performed by human workers at wage w or by AI systems at cost c < w. Integration frictions—technical challenges, organizational adaptation costs, and quality-control requirements—make marginal tasks progressively more difficult to automate, creating smooth trade-offs rather than all-or-nothing adoption. The firm's automation decision becomes choosing the extensive margin: which fraction of tasks to delegate to AI versus retaining human workers.

State of Adoption and Displacement Patterns

Empirical evidence through early 2026 reveals concentrated displacement in specific occupational categories:

• Customer support and service roles: Agentic AI systems handle routine inquiries, troubleshooting, and complaint resolution with minimal human oversight. Salesforce's replacement of 4,000 support agents represents the most publicized case, but adoption spans industries from telecommunications to financial services (CNBC, 2025c).

• Software development and maintenance: AI-assisted coding tools and autonomous agents reduce team size requirements. Cognition's Devin enables productivity multipliers—one engineer replacing five—primarily in routine development tasks, code review, and debugging (CNBC, 2025a).

• Content generation and creative services: Large language models produce marketing copy, technical documentation, and basic creative content, displacing junior writers and content coordinators.

• Data processing and middle management: Automated reporting, analysis, and workflow coordination compress organizational hierarchies, particularly affecting mid-level roles that previously coordinated information flows.

These patterns align with theoretical predictions: AI first substitutes for tasks with well-defined inputs, codifiable procedures, and measurable outputs (Autor et al., 2003). Entry-level positions disproportionately comprise such tasks, creating bottlenecks in traditional career progression pathways (Brynjolfsson et al., 2025a).

Income replacement—the fraction η of displaced wage income recovered through reemployment, transfers, or alternative sources—remains incomplete. Jacobson et al. (1993) document that displaced workers in manufacturing contexts suffer persistent earnings losses averaging 25% even after reemployment. Technology-sector displacements through 2026 have not yet produced sufficient longitudinal data for comparable estimates, but short-run evidence suggests many displaced workers face extended job search periods with downward wage adjustments upon reentry. Absent comprehensive retraining programs or robust income support, significant purchasing power disappears from the economy with each automation wave.

Organizational and Individual Consequences of Competitive Over-Automation

Organizational Performance Impacts

The theoretical model developed by Hemenway Falk and Tsoukalas (2026) predicts that competitive over-automation generates measurable profit erosion even as individual firms perceive positive returns to their own automation decisions. This seemingly paradoxical outcome arises from a strategic externality: when firm i automates, it reduces aggregate demand by displacing workers whose spending would have benefited all N firms in the sector. Under competitive pricing, firm i bears only 1/N of this demand loss while capturing the full cost saving. The remaining (N−1)/N share falls on competitors, creating a classic tragedy of the commons in which individually rational decisions produce collectively suboptimal outcomes.

Quantitatively, the model demonstrates that the over-automation wedge—the gap between equilibrium and cooperative automation rates—equals ℓ(1 − 1/N)/k, where ℓ represents demand loss per automated task, k captures integration friction, and N denotes the number of competitors. For illustrative parameters consistent with observed displacement patterns (c/w = 0.30, λ = 0.5, η = 0.30, N → ∞), the equilibrium automation rate reaches twice the level that would maximize aggregate profit. This wedge increases strictly with market fragmentation: duopolists over-automate modestly, while firms in highly competitive sectors exhibit the widest gaps.

The critical implication for organizational performance is that profit erosion intensifies precisely where firms perceive strongest competitive pressure to automate. Consider a fragmented industry—say, business process outsourcing or regional software development—where dozens of firms compete on cost efficiency. Each observes rivals adopting AI systems and calculates that falling behind technologically would sacrifice market share. The automation arms race ensues not from irrationality but from dominant strategy logic: automating yields higher profits than not automating regardless of what competitors choose, given that each firm bears only a small fraction of the demand destruction its decision creates. Yet when all firms follow this logic simultaneously, the resulting demand contraction reduces every firm's revenue base, leaving all worse off than had they collectively exercised restraint (Hemenway Falk & Tsoukalas, 2026).

Acme Business Solutions (disguised name for confidentiality), a mid-sized U.S.-based provider of back-office services for healthcare administration, illustrates the mechanism. Between 2024 and 2025, Acme automated approximately 40% of its claims-processing tasks using specialized AI models, reducing per-unit labor costs by 35% and initially expanding operating margins. Within six months, however, three major competitors—each serving overlapping client bases—announced comparable automation initiatives with similar displacement rates. By late 2025, Acme's management observed declining contract renewal rates and price pressure as displaced workers, many of whom had patronized clients' services, reduced their healthcare utilization. The aggregate spending contraction in Acme's service region reduced demand for healthcare administration overall. While Acme's cost structure improved, revenue declined faster, compressing margins below pre-automation levels. Management faced a binding constraint: reversing automation unilaterally would worsen cost competitiveness without restoring lost demand, since competitors' displacement decisions remained unchanged.

Individual Wellbeing and Stakeholder Impacts

Workers bear the immediate and most visible costs of displacement. When automation eliminates positions faster than the economy creates comparable alternatives, displaced individuals experience income loss, skill obsolescence, and career disruption. Historical evidence consistently documents persistent negative earnings impacts: Jacobson et al. (1993) found that manufacturing workers displaced during the 1980s recession suffered long-term earnings reductions averaging 25% even after reemployment, with losses concentrated among mid-career workers who had accumulated firm- and industry-specific human capital rendered obsolete by technological change.

AI displacement in white-collar contexts introduces additional dimensions. Unlike manufacturing automation, which historically affected workers with limited geographic mobility and narrow skill transferability, current AI substitution reaches knowledge workers who previously enjoyed labor-market insulation. Customer service representatives, junior software developers, content writers, and data analysts face displacement not due to inadequate skills but because AI systems achieve comparable task performance at lower cost. Reemployment prospects depend critically on whether displaced workers can transition to roles requiring capabilities AI cannot easily replicate—typically involving complex interpersonal interaction, creative judgment, or strategic decision-making under ambiguity (Autor et al., 2003; Acemoglu & Restrepo, 2019).

The income-replacement rate η—the fraction of lost wages recovered through alternative employment, transfers, or household adjustments—determines whether displacement translates into persistent consumption loss. When η < 1, displaced workers reduce spending, and this demand destruction propagates through the economy. Critically, the competitive externality intensifies as η falls: lower replacement means higher per-task demand loss ℓ = λ(1 − η)w, widening the over-automation wedge and amplifying both profit erosion and worker harm (Hemenway Falk & Tsoukalas, 2026).

FintechForward (name disguised), a regional financial technology firm, experienced displacement dynamics that illustrate distributional consequences. In 2025, the company adopted an AI-driven customer relationship management system that automated routine client interactions, reducing its customer-service workforce from 150 to 45 employees. Severance packages provided three months' income replacement, after which displaced workers entered the labor market. Follow-up interviews conducted six months post-displacement revealed that 40% had secured comparable positions (full income replacement, η ≈ 1), 35% found employment in adjacent roles at 15–20% lower compensation (η ≈ 0.80–0.85), and 25% remained underemployed in gig economy or part-time positions (η ≈ 0.40–0.60). Aggregate spending among the displaced cohort, tracked through anonymized transaction data, declined approximately 30% relative to pre-displacement levels, concentrated in discretionary categories—dining, entertainment, and durable goods—that disproportionately support other local firms.

The purchasing-power channel creates feedback loops that harm workers beyond the initially displaced cohort. As spending contracts, downstream firms—restaurants, retailers, service providers patronized by displaced workers—experience revenue declines. Some respond with their own workforce reductions or automation adoption, propagating displacement across sectors. This multiplier effect explains why aggregate consequences of automation can exceed the sum of direct displacements: each automation decision erodes not only the displaced workers' consumption but also reduces demand for complementary labor throughout the economy (Murphy et al., 1989; Rosenstein-Rodan, 1943).

Evidence-Based Organizational Responses

Table 1: Corporate AI Automation and Displacement Case Studies

Organization	Sector	Automation Mechanism	Estimated Job Displacement	Intervention Strategy	Income Replacement Rate (Inferred)	Outcome Summary
TechCorp (Disguised)	Software	AI systems (Tier-1 support)	60% of Tier-1 support interactions	6-month intensive retraining program in AI oversight and data analysis	0.95	75% of participants secured positions at 110% of prior wages; substantially mitigated demand loss compared to baseline.
Technology Services Firm (Disguised)	IT Services	AI-driven development tools	15% of engineering workforce	Capability Security (20% work time allocated to portable skill learning)	1.00	80% of displaced workers secured comparable/superior roles within 4 months due to verified external credentials.
Healthcare Diagnostics Group (Disguised)	Healthcare/Radiology	AI imaging analysis	None (Headcount stable)	Augmentation (Task reallocation/Decision support)	1.00	Diagnostic throughput increased by 35% per radiologist; volume expanded to meet latent demand; purchasing power preserved.
Regional Manufacturing Alliance (Disguised)	Manufacturing	Production automation	800 workers (across 15 firms)	Sectoral Consortium (joint retraining fund, shared career services, inter-firm job matching)	0.85	Prevented competitive "races to the bottom"; achieved higher reemployment rates than isolated firm efforts.
Finance Sector Restructuring Initiative (Disguised)	Banking	Back-office automation	800 workers	6 months severance, 12 months health insurance, $15k tuition reimbursement	0.85	Temporary preservation of consumption during severance; 60% utilized retraining; average reemployment wages fell 15% later.
Industrial Cooperative Network (Disguised)	Manufacturing	Automated machining operations	50 workers	Mandatory equity participation (30% ESOP) and inter-network reemployment	0.75	Internalized a larger share of demand loss than conventional firms, but still faced over-automation incentives relative to a central planner.
FintechForward (Disguised)	Fintech	AI-driven CRM system	105 employees	Severance packages (3 months)	0.70	Workforce reduced from 150 to 45; 25% remained underemployed; aggregate spending of the cohort declined 30%.
Midwest Manufacturing Cooperative (Disguised)	Manufacturing/Fabrication	Robotic welding systems	40% of welding workforce	Transparent communication, advance notification (18 months), worker votes, severance	0.50	High process satisfaction and preserved reputation, but failed to prevent local demand destruction from displaced worker-owners.
Goldman Sachs	Finance/Investment Banking	Cognition's Devin (autonomous coder)	4 out of 5 engineers (per team)	Not in source	0.40	Piloting autonomous coders to enable one senior engineer to perform work previously requiring five-person teams.
Block Inc.	Technology/Fintech	AI capabilities	5,000 positions	Not in source	0.30	Eliminated nearly half of workforce; CEO predicted most companies will follow to render roles obsolete.
Salesforce	Technology/SaaS	Agentic AI systems	4,000 agents	Not in source	0.30	Replaced customer-support agents to reduce headcount; individual productivity gains realized but collective demand impacts cited as risk.
Acme Business Solutions (Disguised)	Healthcare Administration	Specialized AI models	40% of claims-processing tasks	Not in source	0.30	Initial cost reduction of 35% was offset by declining revenue as displacement reduced regional healthcare utilization; margins fell below pre-automation levels.

Organizations facing AI adoption decisions confront a dual challenge: capturing efficiency gains from automation while managing displacement consequences that feed back into demand conditions. This section examines five categories of organizational response, evaluating their effectiveness against the structural externality identified above and drawing on both theoretical analysis and practitioner-oriented case evidence.

Transparent Communication and Procedural Justice

Organizational justice research demonstrates that displacement decisions perceived as procedurally fair—characterized by advance notice, transparent rationale, and opportunities for affected employees to voice concerns—mitigate adverse psychological and community-reputation consequences (Brockner et al., 1994; Folger & Cropanzano, 1998). Procedural fairness does not, however, alter the economic incentive to automate or the magnitude of demand destruction resulting from displacement.

Firms implementing transparent communication strategies typically adopt multi-phase approaches:

• Advance notification: Announcing automation plans 6–12 months prior to implementation, providing workers time to pursue retraining or job search while still employed.

• Rationale disclosure: Explaining competitive pressures, cost structures, and technological capabilities that necessitate automation, contextualizing decisions within industry-wide trends rather than attributing blame to individual performance.

• Input solicitation: Creating forums—town halls, focus groups, anonymous surveys—where affected workers can express concerns, propose alternatives, and influence implementation details (timing, support services, severance terms).

• Severance and transition support: Offering financial packages extending beyond statutory minimums, coupled with outplacement services, career counseling, and access to retraining programs.

Midwest Manufacturing Cooperative (name disguised), a worker-owned fabrication firm, implemented an exemplary transparent approach when adopting robotic welding systems in 2024. Management held quarterly informational sessions beginning 18 months before deployment, explaining technological capabilities, cost projections, and competitive necessity. Worker-owners voted on implementation timelines and severance terms through cooperative governance structures. Despite ultimately reducing the welding workforce by 40%, post-implementation surveys indicated high satisfaction with process fairness, and the firm retained strong community reputation. Critically, however, the cooperative's procedural justice approach did not prevent the demand-side externality: displaced worker-owners reduced local consumption, affecting other cooperative members' businesses in the regional economy.

Mechanisms and Limitations: Transparent communication operates primarily on within-firm outcomes—worker morale, organizational reputation, legal risk mitigation—rather than correcting the cross-firm externality. A firm that communicates transparently still captures full cost savings while bearing only 1/N of aggregate demand loss. Procedural justice may reduce transitional friction and preserve employer brand, facilitating future hiring and stakeholder relations, but it does not change the per-task automation calculus that drives over-adoption. The structural externality persists regardless of communication quality.

Upskilling, Retraining, and Capability Building

Human capital investments that raise displaced workers' income-replacement rate η—through retraining programs enabling transitions to higher-wage roles, or upskilling initiatives that prepare incumbent workers for AI-augmented rather than AI-substituted positions—directly shrink the demand-loss parameter ℓ = λ(1 − η)w and narrow the over-automation wedge (Hemenway Falk & Tsoukalas, 2026). This mechanism operates on the right margin, unlike communication strategies, but cannot fully eliminate the externality unless η reaches or exceeds unity (complete income replacement).

Effective retraining initiatives share common features:

• Skills gap analysis: Identifying capabilities required for emerging roles—typically involving judgment, creativity, complex communication, or oversight of AI systems—that automation cannot easily replicate.

• Tailored curriculum: Designing programs that build on workers' existing knowledge bases rather than requiring wholesale skill replacement, reducing time-to-reemployment and increasing completion rates.

• Credentialing partnerships: Collaborating with educational institutions, industry associations, or certification bodies to ensure training yields recognized, portable credentials that facilitate external labor-market mobility.

• Income support during transition: Providing wage continuation or stipends that enable workers to participate in full-time training without immediate financial distress.

TechCorp Training Initiative (name disguised), implemented by a large software firm following customer-service automation in 2025, illustrates structured upskilling. After AI systems automated 60% of Tier-1 support interactions, TechCorp offered displaced representatives a six-month intensive program in data analysis, AI system oversight, and advanced customer success management—roles requiring human judgment to interpret AI outputs, handle complex escalations, and manage high-value client relationships. The program combined technical instruction with business communication training and culminated in industry-recognized certifications. Of 200 participants, 75% secured positions internally or externally at wages averaging 110% of prior compensation (η ≈ 1.10), 15% found comparable roles (η ≈ 1.0), and 10% exited the labor force or accepted lower-wage alternatives. Aggregate income replacement across the cohort reached approximately η ≈ 0.95, substantially mitigating demand loss relative to scenarios with minimal retraining support.

Mechanisms and Limitations: Raising η reduces ℓ, which narrows the wedge ℓ(1 − 1/N)/k but cannot eliminate it entirely unless retraining achieves η ≥ 1—displaced workers secure equal or higher wages, fully offsetting lost purchasing power. Achieving η ≥ 1 economy-wide requires either exceptionally successful retraining (displaced workers transition to systematically higher-productivity roles) or rapid creation of new high-wage occupations (Acemoglu & Restrepo, 2019). Historical evidence offers mixed support: some technological transitions eventually produce net employment and wage gains (the rise of personal computing created vast new occupational categories in software development, IT management, and digital services), while others generate persistent displacement without comparable job creation in displaced workers' skill neighborhoods (manufacturing automation during the 1980s and 1990s concentrated losses among mid-skill workers who faced limited reemployment prospects in comparable-wage roles).

Critically, even successful firm-level upskilling does not address the strategic externality. If TechCorp's retraining raises η to 0.95 for its displaced workers, the effective demand loss per task falls from ℓ = λ(1 − 0.30)w ≈ 0.35w (assuming baseline η = 0.30) to ℓ ≈ 0.025w, substantially reducing the wedge. However, competitors who do not invest comparably in retraining still impose higher demand losses, and no individual firm internalizes the cross-firm spillovers from its retraining investments. The uncoordinated equilibrium therefore under-provides retraining from a social perspective, even though retraining addresses the right margin by raising η.

Worker Equity Participation and Profit-Sharing

When workers hold equity stakes in firms or receive shares of automation-generated profits, displaced wage income is partially offset by capital income, raising the effective replacement rate. Hemenway Falk and Tsoukalas (2026) demonstrate that profit-sharing at rate ϵ ∈ [0,1] reduces the over-automation wedge to ℓ(1 − λϵ)(N − 1)/[k(N − λϵ(N − 1))], which is strictly smaller than the baseline wedge but remains positive for all ϵ < 1/λ. When workers' marginal propensity to consume in the sector (λ) is less than one—meaning each dollar of profit-sharing generates less than one dollar of sectoral demand—even full profit-sharing (ϵ = 1) cannot eliminate the externality.

Profit-sharing structures take multiple forms:

• Employee stock ownership plans (ESOPs): Workers acquire equity stakes, typically through employer contributions to retirement accounts or direct stock grants. Dividend income and capital appreciation provide income streams that partially replace wages post-displacement.

• Collective profit-sharing agreements: Annual or quarterly bonuses tied to firm profitability, distributed proportionally to compensation or tenure, generating income that continues if workers retain employment but disappears upon separation.

• Equity grants with vesting: Options or restricted stock units that vest over multi-year horizons, creating deferred compensation that may partially offset near-term displacement.

• Cooperative ownership structures: Worker-owned firms in which members hold equal equity stakes and governance rights, ensuring displaced workers retain capital income proportional to their ownership shares.

Industrial Cooperative Network (name disguised), a federation of worker-owned manufacturing firms in the U.S. Northeast, maintains mandatory equity participation: all employees acquire ownership stakes equal to approximately 30% of annual compensation over a five-year vesting period. When a member firm automated machining operations in 2024–2025, displacing 50 workers, those individuals retained equity positions generating dividend income averaging 15% of prior wages. Combined with reemployment at other cooperative network firms (further 60% income replacement), aggregate replacement reached η ≈ 0.75, substantially above typical displacement scenarios but still below full replacement. The cooperative structure internalized a larger share of demand loss than conventional firms—worker-owners' profit-sharing effectively increased the divisor in the wedge formula from N to N − λϵ(N − 1)—but did not eliminate over-automation incentives relative to what a network-wide planner optimizing over all cooperatives jointly would choose.

Mechanisms and Limitations: Profit-sharing operates on the demand-loss parameter by recycling capital income (which would otherwise exhibit low marginal propensity to consume in the sector) back into worker consumption. Each dollar of profit shared with workers generates λ dollars of sectoral demand, where λ ∈ (0,1] represents workers' sectoral spending propensity. This mechanism narrows the effective number of competitors from N to Nϵ = N − λϵ(N − 1), reducing the uninternalized fraction of demand loss. However, unless λ = 1 (workers spend all income exclusively in the sector—an implausible extreme), the wedge remains strictly positive even at ϵ = 1. Structural limits arise because the externality is multilateral: each firm's automation harms all N competitors, and bilateral profit-sharing between a firm and its workers cannot compensate for demand losses falling on rivals' workers (Hemenway Falk & Tsoukalas, 2026).

A distinct question is whether profit-sharing emerges voluntarily. Hemenway Falk and Tsoukalas (2026) demonstrate that profit-sharing is not a dominant strategy: each firm captures only 1/N of the demand benefit from sharing profits with its workers, while bearing the full cost (reduced retained earnings). Consequently, ϵ = 0 dominates any ϵ > 0 in the non-cooperative game, meaning profit-sharing requires either regulatory mandate or collective bargaining agreements that bind all firms simultaneously—precisely the coordination failures the externality creates in automation decisions themselves.

Operating Model Adaptation: Augmentation Over Substitution

AI deployment choices exist along a continuum from pure substitution (AI replaces human workers entirely, as in automated customer service bots) to pure augmentation (AI enhances human productivity without displacement, as in diagnostic assistance for radiologists). Augmentation strategies reduce gross displacement per unit of AI adoption, effectively lowering the rate at which automation converts into lost jobs and purchasing power. However, augmentation does not alter the strategic incentive to automate beyond collectively optimal levels when substitution opportunities exist.

Augmentation-focused operating models emphasize:

• Task reallocation: Assigning routine, codifiable sub-tasks to AI systems while reserving judgment-intensive, interpersonal, or creative components for human workers. For example, AI drafts initial content, and humans refine, contextualize, and ensure quality.

• Decision support rather than decision replacement: Implementing AI as advisory tools—providing analysis, recommendations, scenario modeling—while retaining humans in final decision-making roles.

• Human-in-the-loop systems: Designing workflows requiring human oversight, approval, or exception-handling at critical junctures, ensuring AI operates under continuous supervision rather than autonomously.

• Skill upgrading within roles: Redefining job responsibilities to emphasize capabilities AI cannot easily replicate (complex communication, strategic thinking, ethical judgment), with AI handling lower-value components that previously consumed workers' time.

Healthcare Diagnostics Group (name disguised), a network of radiology practices, adopted an augmentation-first approach when integrating AI imaging analysis in 2024. Rather than replacing radiologists, the group deployed AI systems to pre-screen scans, flag potential abnormalities, and prioritize case queues. Radiologists reviewed AI outputs, confirmed or corrected diagnoses, and handled complex cases requiring contextual medical judgment. The augmentation increased diagnostic throughput per radiologist by approximately 35% without reducing headcount; instead, the practice expanded patient volume to meet latent demand previously constrained by radiologist availability. Compensation remained stable, preserving purchasing power.

Mechanisms and Limitations: Augmentation reduces displacement per unit of AI adoption but does not eliminate pressure toward substitution when substitution is technologically feasible and cost-advantageous. If AI capabilities improve to the point where substitution becomes viable—e.g., diagnostic AI achieves performance parity with human radiologists even on complex cases—the competitive incentive to fully automate resurfaces. Firms that persist with augmentation while competitors substitute face cost disadvantages that erode market position (Bastani & Cachon, 2025).

Moreover, augmentation strategies do not address the core externality mechanism. Even if a firm chooses augmentation and avoids displacement, competitors who substitute workers still generate demand losses that reduce the augmenting firm's revenue. The externality operates at the sector level: any displacement anywhere in the competitive system imposes uninternalized costs on all firms. Augmentation is therefore better characterized as a damage mitigation strategy—reducing a firm's own contribution to aggregate displacement—rather than a solution to the structural over-automation problem.

Empirical evidence on augmentation's sustainability remains mixed. Brynjolfsson et al. (2025b) document productivity gains from AI-augmented customer service but note that firms retaining larger human workforces face cost pressures when competitors adopt more substitution-heavy approaches. As AI reliability improves, the returns to human oversight decline, weakening economic justification for keeping humans in the loop (Bastani & Cachon, 2025). Market dynamics may therefore push firms toward substitution over time even when augmentation initially predominates.

Financial and Benefit Support: Income Continuity Programs

Severance packages, extended health benefits, wage insurance, and supplemental unemployment assistance raise the income-replacement rate η by providing temporary income continuity during job search or retraining. Unlike retraining programs (which raise η by facilitating transitions to comparable-wage roles), income support raises η by directly replacing lost wages over finite horizons. This reduces immediate consumption disruption but does not build durable human capital or create pathways to sustainable reemployment.

Income support programs include:

• Enhanced severance: Cash payments calibrated to tenure, seniority, or wage levels, often extending 3–12 months beyond statutory requirements. Enables continued consumption during job search without immediate financial distress.

• Wage insurance: Supplemental payments covering a fraction (e.g., 50%) of the wage gap if displaced workers accept lower-paying reemployment. Reduces income loss from downward occupational mobility.

• Benefit continuation: Extending health insurance, retirement contributions, or other non-wage benefits beyond termination dates, reducing non-wage income loss.

• Unemployment insurance enhancement: Employer-funded supplements to public unemployment systems, raising replacement rates or extending eligibility durations.

Finance Sector Restructuring Initiative (name disguised), implemented by a large commercial bank following back-office automation in 2025, combined severance (six months' base salary), health insurance continuation (12 months), and tuition reimbursement for credentialed retraining programs (up to $15,000). Of 800 displaced workers, approximately 60% utilized retraining support, and extended severance enabled job searches averaging four months without severe financial distress. Aggregate income replacement during the severance window reached η ≈ 0.85–0.90 (severance plus partial reemployment earnings for early re-entrants), falling to η ≈ 0.65 post-severance as reemployment wages averaged 15% below prior levels.

Mechanisms and Limitations: Income support operates identically to retraining in the theoretical framework—both raise η, thereby reducing ℓ = λ(1 − η)w and narrowing the over-automation wedge—but differ in durability. Severance provides temporary income replacement, preserving consumption over a finite window but creating a "cliff" when benefits exhaust. If reemployment at comparable wages does not occur before severance expires, η falls sharply, and demand destruction resumes. Retraining, by contrast, invests in human capital that raises long-run reemployment wages, sustaining higher η indefinitely.

The externality persists because, as with retraining, no individual firm internalizes the demand benefits of raising η for its displaced workers beyond the 1/N share accruing to itself. Generous severance is therefore undersupplied in competitive equilibrium: firms bear full costs (severance payments) while capturing only fractional demand benefits. Collective action problems prevent firms from coordinating on sector-wide income support standards that would internalize cross-firm spillovers.

Policy implications suggest that income support is most effective as a complement to retraining rather than a substitute. Severance buys time for workers to complete skill-building programs without financial distress, raising program completion rates and ultimate reemployment success. Wage insurance reduces the penalty for accepting transitional positions in adjacent occupations, facilitating occupational mobility. However, income support absent retraining pathways merely postpones rather than prevents consumption loss, achieving temporary demand preservation at the cost of permanent human capital under-investment.

Building Long-Term Organizational and Sectoral Resilience

The organizational responses examined above—communication strategies, retraining, profit-sharing, augmentation, and income support—mitigate displacement consequences at firm or worker levels but do not eliminate the structural externality driving competitive over-automation. This section examines three systemic strategies for building resilience: recalibrating the psychological contract between firms and workers, developing distributed governance structures that internalize cross-firm spillovers, and establishing continuous learning ecosystems that sustain high income-replacement rates even amid ongoing technological change.

Psychological Contract Recalibration: From Employment Security to Capability Security

The traditional employment relationship—implicit expectations of long-term tenure in exchange for loyalty and performance—erodes under rapid automation (Rousseau, 1995). Workers who anticipate displacement irrespective of individual performance disengage, reducing productivity and organizational commitment. Forward-looking firms recalibrate the psychological contract from employment security (expectation of retaining a specific role indefinitely) to capability security (expectation of maintaining employability through continuous skill development).

Capability-security contracts emphasize:

• Portable skill development: Training investments that build general human capital—capabilities transferable across employers and occupations—rather than firm-specific skills rendered obsolete when automation eliminates particular roles.

• Transparent career pathing: Communicating which roles are automation-vulnerable versus automation-resistant, enabling workers to pursue internal mobility toward positions less susceptible to displacement.

• Learning time allocation: Dedicating work hours explicitly to skill-building, ensuring that learning competes with production on equal footing rather than requiring off-hours personal investment.

• Credentialing partnerships: Facilitating external certifications, degrees, or professional credentials that signal capabilities to the external labor market, raising reemployment prospects post-displacement.

Technology Services Firm (name disguised), facing predictable displacement from AI-driven development tools, implemented a capability-security model in 2024. The firm allocated 20% of each engineer's work time to learning activities—courses, certifications, cross-functional projects—selected collaboratively between employees and managers based on individual career goals and emerging technology trends. The firm subsidized external certifications and maintained partnerships with universities offering accelerated degree programs. When automation reduced routine coding demand in 2025, displacing approximately 15% of the engineering workforce, affected individuals exited with portfolios of verified skills, external credentials, and clear narratives of capability development. Post-displacement surveys indicated 80% secured comparable or superior positions within four months (η ≈ 1.0), well above sector averages.

This approach raises η by preparing workers for reemployment before displacement occurs, reducing frictional unemployment and skill obsolescence. However, it does not alter the automation incentive itself: firms still automate beyond collectively optimal levels because each bears only 1/N of demand loss. Psychological contract recalibration is therefore best understood as a worker-protection mechanism that softens displacement consequences rather than a market-failure correction that prevents excessive automation.

Distributed Governance: Industry Consortia and Sectoral Bargaining

Centralized coordination mechanisms—industry associations, sectoral collective bargaining agreements, or employer consortia—can internalize cross-firm externalities by setting automation standards or displacement-mitigation requirements that bind all participants simultaneously. Hemenway Falk and Tsoukalas (2026) demonstrate that partial coalitions of M < N firms choosing automation rates jointly achieve α^M = (s − Mℓ/N)/k, internalizing M/N of the aggregate demand loss but still over-automating relative to the full cooperative optimum unless M = N (the grand coalition). Voluntary coalitions face incentive-compatibility failures: because automation is a dominant strategy, defection from any agreement remains individually profitable, preventing stable coalitions from forming without binding enforcement.

Nonetheless, distributed governance structures can facilitate coordination:

• Industry-wide automation standards: Trade associations or sectoral bodies establishing guidelines on displacement notification, severance minimums, retraining investments, or income-replacement rates. Standards internalize spillovers by committing all members to comparable practices, reducing competitive disadvantage from unilateral generosity.

• Sectoral collective bargaining: Unions negotiating with employer federations to establish automation protocols, worker transition rights, and profit-sharing arrangements that apply uniformly across all firms within a sector or region. Uniform application prevents races to the bottom.

• Shared retraining infrastructure: Consortia pooling resources to fund industry-wide retraining programs, spreading costs across multiple firms and ensuring that training investments raise η for the entire sectoral workforce rather than only benefiting individual firms' displaced workers.

• Demand-preservation agreements: Explicit compacts among competitors to limit aggregate displacement rates—e.g., committing to maintain workforce levels at X% of baseline for Y years—with monitoring and penalty mechanisms enforcing compliance.

Regional Manufacturing Alliance (name disguised), comprising 15 mid-sized fabrication firms in the U.S. Midwest, formed a sectoral consortium in 2023 to coordinate automation adoption and worker transitions. The alliance established common standards: minimum 6-month severance, joint retraining fund (1.5% of payroll from all members), shared career counseling services, and displacement notification to a centralized clearinghouse enabling inter-firm job matching. When individual firms automated production processes during 2024–2025, displaced workers accessed consortium-wide reemployment opportunities and standardized retraining pathways. Aggregate displacement across the 15 firms reached approximately 800 workers, but consortium-facilitated reemployment (within the alliance or through retraining pathways) achieved η ≈ 0.85, substantially above what isolated firm-level efforts typically accomplish. Moreover, uniform standards prevented competitive leapfrogging—no firm gained cost advantage by shirking severance or retraining obligations.

Mechanisms and Limitations: Industry consortia effectively increase the size of the cooperating unit from individual firms to coalitions, raising the internalized share of demand loss from 1/N to M/N. When M encompasses most firms in a sector, the residual externality becomes small. However, enforceability remains critical: voluntary consortia face defection risk (firms exiting to avoid obligations when automation advantages outweigh consortium membership benefits), and legal constraints in some jurisdictions restrict industry-wide agreements that might constitute anti-competitive coordination. Additionally, consortia operate at sectoral scales and cannot address cross-sector spillovers—displacement in manufacturing reduces demand for services, and vice versa—leaving macroeconomic externalities unaddressed even when sectoral coordination succeeds.

Continuous Learning Ecosystems: Sustaining High η Through Adaptive Education

Long-term resilience requires sustaining high income-replacement rates (η) across successive waves of technological change, not just responding to individual displacement events. This necessitates embedding continuous learning into economic structures—education systems that adapt curricula rapidly to emerging skill demands, credentialing mechanisms that validate learning regardless of institutional source, and financing models that support mid-career transitions without prohibitive costs.

Effective continuous learning ecosystems incorporate:

• Modular, stackable credentials: Micro-credentials, certificates, and badges that recognize specific competencies, enabling workers to build qualifications incrementally rather than requiring multi-year degree programs. Facilitates continuous upskilling while employed.

• Industry-responsive curriculum design: Educational institutions maintaining advisory partnerships with employers, updating course content based on real-time labor-market signals about emerging skill gaps and obsolescing capabilities.

• Portable financing mechanisms: Income-share agreements, lifelong learning accounts, or publicly subsidized retraining funds that enable workers to finance education without employer-specific commitments, preserving mobility and adaptability.

• Recognition of prior learning: Credit systems that assess and credential skills acquired through work experience, non-formal training, or self-directed learning, reducing redundancy and accelerating credential completion for displaced workers.

National Retraining Partnership (based on real-world examples in Nordic countries), a public-private collaboration in a small European economy, exemplifies systemic continuous learning. The partnership links employers, unions, educational institutions, and government agencies in co-designing responsive training programs. Workers access individualized learning accounts (public funding + employer contributions) that cover tuition, income replacement during full-time study, and credentialing costs. Modular curricula enable workers to pursue micro-credentials aligned with immediate job opportunities or stack credentials toward degrees over extended timelines. Labor-market information systems provide real-time data on occupation-specific demand, guiding workers toward high-value training pathways. Empirical evaluations (not cited due to confidentiality) suggest that displaced workers participating in this ecosystem achieve η ≈ 0.90–0.95 within 12–18 months, substantially higher than uncoordinated retraining efforts.

Mechanisms and Limitations: Continuous learning ecosystems raise η durably by reducing the time, cost, and risk associated with skill transitions. When workers anticipate accessible retraining pathways, displacement becomes less catastrophic, and aggregate consumption volatility diminishes. However, ecosystem development requires substantial public investment, cross-sector coordination, and political commitment that may not emerge voluntarily from competitive market dynamics. Moreover, even perfectly functioning ecosystems raise η toward but not necessarily beyond unity—unless retraining systematically places workers in higher-wage roles (η > 1), the demand-loss parameter ℓ remains positive, and over-automation persists. Finally, ecosystems do not address the incentive margin: firms still perceive dominant strategies to automate beyond collectively optimal rates because demand-loss internalization depends on N, not on η.

Conclusion

The analysis presented here demonstrates that AI-driven labor displacement in competitive markets generates a structural externality that voluntary organizational responses cannot fully eliminate. Each firm captures complete cost savings from automation while bearing only a fractional share of the resulting demand destruction, creating dominant-strategy incentives to automate beyond collectively efficient levels. This over-automation harms both workers—through direct income loss—and firm owners—through profit erosion from aggregate demand contraction—making it a deadweight loss rather than a mere distributional conflict.

The evidence-based organizational responses examined—transparent communication, retraining investments, profit-sharing arrangements, augmentation strategies, and income support programs—mitigate displacement consequences and narrow the over-automation wedge by raising income-replacement rates (η) or reducing substitution intensity. However, none eliminate the externality entirely because none internalize the cross-firm demand spillovers that drive excessive automation. Even comprehensive approaches combining multiple interventions leave residual over-automation unless income replacement reaches or exceeds 100% economy-wide—a threshold historical evidence suggests is rarely achieved (Jacobson et al., 1993; Autor et al., 2024).

Three actionable implications emerge for practitioners:

1. Retraining investments yield highest returns: Among voluntary organizational responses, capability-building programs that raise η operate on the correct economic margin—reducing per-task demand loss—and generate measurable reductions in the over-automation wedge (Hemenway Falk & Tsoukalas, 2026). Firms facing automation decisions should prioritize human capital investments that facilitate displaced workers' transitions to comparable-wage roles, recognizing that higher η benefits all firms in the sector by preserving aggregate purchasing power.

2. Industry coordination amplifies effectiveness: Sectoral consortia or collective bargaining agreements that establish uniform displacement standards internalize larger shares of demand externalities than isolated firm-level initiatives. Where feasible, cooperative approaches—shared retraining infrastructure, common severance minimums, inter-firm job-matching systems—achieve superior outcomes by preventing competitive races toward minimal worker support.

3. Policy intervention remains necessary: Because the externality arises from competitive market structure rather than information failures or transitional frictions, market-based solutions alone cannot restore efficiency. Organizations should engage constructively with policy frameworks—automation taxation, public retraining systems, or regulatory standards—that correct the per-task automation incentive rather than opposing intervention as market interference.

For policymakers, the findings suggest that conventional labor-market interventions—unemployment insurance, job search assistance, or income support—address displacement consequences but not the competitive incentives driving excessive automation. Correcting the structural distortion requires instruments operating on the automation decision margin itself, with Pigouvian automation taxes emerging as the theoretically optimal solution (Hemenway Falk & Tsoukalas, 2026). Tax revenue can fund retraining programs that raise η, creating a virtuous cycle: higher income replacement reduces the demand externality, shrinking the required tax rate over time and potentially rendering the intervention self-limiting as labor markets adapt.

The broader implication is that AI displacement discourse should shift from ex post adjustment assistance toward ex ante incentive correction. The competitive trap—rational firms over-automating despite collective harm—is not inevitable but results from market failures that policy can address. Recognizing that both workers and firm owners lose from excessive automation reframes intervention not as redistribution favoring one factor class over another, but as efficiency enhancement benefiting all stakeholders. This perspective may enable more productive dialogue between business, labor, and government constituencies than distributional framings that position interests as inherently opposed.

Research Infographic

The AI Automation Trap II Slide Deck

References

1. Acemoglu, D. (2025). The simple macroeconomics of AI. Economic Policy, 40(121), 13–58.

2. 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.

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

4. Autor, D. H., Chin, C., Salomons, A., & Seegmiller, B. (2024). New frontiers: The origins and content of new work, 1940–2018. Quarterly Journal of Economics, 139(3), 1399–1465.

5. Autor, D. H., Levy, F., & Murnane, R. J. (2003). The skill content of recent technological change: An empirical exploration. Quarterly Journal of Economics, 118(4), 1279–1333.

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

7. Brockner, J., Konovsky, M., Cooper-Schneider, R., Folger, R., Martin, C., & Bies, R. J. (1994). Interactive effects of procedural justice and outcome negativity on victims and survivors of job loss. Academy of Management Journal, 37(2), 397–409.

8. 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.

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

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

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

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

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

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

15. Folger, R., & Cropanzano, R. (1998). Organizational justice and human resource management. Sage Publications.

16. Hemenway Falk, B., & Tsoukalas, G. (2026). The AI layoff trap. arXiv preprint arXiv:2603.20617.

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

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

19. Keynes, J. M. (1930). Economic possibilities for our grandchildren. In Essays in Persuasion (pp. 358–373). Macmillan.

20. 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.

21. Murphy, K. M., Shleifer, A., & Vishny, R. W. (1989). Industrialization and the big push. Journal of Political Economy, 97(5), 1003–1026.

22. Ricardo, D. (1821). On the principles of political economy and taxation (3rd ed.). John Murray.

23. Rosenstein-Rodan, P. N. (1943). Problems of industrialisation of Eastern and South-Eastern Europe. Economic Journal, 53(210/211), 202–211.

24. Rousseau, D. M. (1995). Psychological contracts in organizations: Understanding written and unwritten agreements. Sage Publications.

25. 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 Competitive Trap: How AI-Driven Automation Creates Collective Market Failure. Human Capital Leadership Review, 36(1). doi.org/10.70175/hclreview.2020.36.1.7