The Remote Work–AI Paradox: Rethinking the Decline in Early-Career Hiring

Listen to a review of this article:

Abstract: Recent evidence shows significant declines in early-career hiring across advanced economies since 2022, prompting urgent questions about workforce development and productivity. While emerging research attempts to isolate generative AI as the primary driver, the relationship between technological change, organizational structure, and junior talent acquisition remains poorly understood. This analysis critically examines a recent working paper by Lambert and Schindler (2026) that challenges AI-centric narratives by highlighting remote work as a competing explanation. Drawing on labor economics and organizational behavior research, we argue that univariate explanations oversimplify a multifaceted phenomenon involving measurement challenges, correlated exposures, and context-dependent mechanisms. The evidence suggests both forces likely operate simultaneously through distinct channels—AI through task automation and skill polarization, remote work through supervision costs and learning friction—with their relative importance varying by occupation, firm capability, and implementation approach. Practitioners and policymakers require more nuanced frameworks that acknowledge uncertainty, emphasize organizational adaptation, and avoid premature dismissal of either explanation.

Something troubling is happening in graduate hiring. According to a new working paper by Lambert and Schindler (2026), across the United States, United Kingdom, Canada, and Australia, the share of new positions filled by early-career workers has fallen 8–11 percentage points below pre-pandemic levels. For context, this represents roughly one in every nine entry-level opportunities shifting to experienced candidates. The consequences extend beyond immediate employment statistics: early-career roles serve as the primary mechanism for human capital formation, organizational renewal, and intergenerational economic mobility. The classic human capital literature emphasizes that skills developed during initial work experiences shape lifetime earnings trajectories (Arrow, 1962; Mincer, 1974). Persistent contraction threatens long-run productivity growth while imposing concentrated costs on emerging professionals.

Two explanations dominate current discourse. The first centers on generative AI—particularly large language models like ChatGPT—which achieved mainstream adoption in late 2022. Proponents argue these tools now perform cognitive and analytical tasks historically delegated to junior staff, fundamentally restructuring demand for inexperienced labor. Multiple recent studies document correlations between AI exposure and declining junior employment (Brynjolfsson et al., 2025a; Teeselink, 2025; Azar et al., 2025; Hosseini Maasoum and Lichtinger, 2025).

The second explanation focuses on remote and hybrid work arrangements, which persisted well above pre-pandemic levels and may create organizational friction around supervision, mentorship, and skill development that disproportionately affects junior workers. Research suggests that remote work can reduce collaboration intensity, slow learning, and complicate performance evaluation—all challenges that may fall more heavily on inexperienced workers (Yang et al., 2022; Emanuel et al., 2024; Aksoy et al., 2026).

The Lambert and Schindler (2026) working paper challenges the AI-centric narrative, presenting evidence that remote work exposure better predicts junior hiring declines than AI exposure when both factors enter statistical models jointly. Some have interpreted this as exonerating AI from responsibility for early-career displacement. We contend this interpretation overstates the findings and reflects broader analytical challenges that merit careful examination.

The Early-Career Hiring Landscape

Defining "Junior" and "Early-Career" in Contemporary Labor Markets

Operational definitions matter enormously when measuring workforce composition changes. Lambert and Schindler classify workers using resume-derived seniority levels (combining "entry-level" and "junior" categories) and job postings requiring ≤3 years experience. These approaches capture intuitive concepts but introduce measurement considerations worth noting.

Resume-based classification relies on algorithmic parsing of job titles and career trajectories, which may systematically differ across demographic groups, industries, and occupational structures. A 25-year-old software engineer with two prior roles occupies a different position than a 25-year-old management consultant with equivalent tenure, yet both might be coded identically. Job posting requirements reflect employer stated preferences, which research suggests can diverge from actual hiring behavior—particularly regarding experience thresholds that often serve as negotiable signals rather than binding constraints.

Alternative approaches might include age-based definitions (workers under 25 or 30), tenure-based measures (first three years in labor force), or wage-based classification (bottom quartile of occupation-specific wage distributions). Each operationalization could yield different estimates and potentially different substantive conclusions. The Lambert and Schindler approach is reasonable but not uniquely "correct."

State of Practice: The Post-2022 Contraction

The empirical phenomenon itself appears robust across their data sources. Lambert and Schindler analyze 243 million new employer-employee matches from resume data and 407 million online job vacancy postings across four countries during 2017-2025. Their data show sharp declines in junior-to-senior ratios beginning in late 2022, coinciding with both generative AI's public breakthrough and the stabilization of hybrid work norms.

Three stylized facts from their analysis characterize the shift. First, the decline concentrates in white-collar, knowledge-intensive occupations with high computer use—precisely the roles exposed to both AI capabilities and remote work feasibility. Second, the contraction appears in both hiring flow measures (new positions filled) and recruitment demand signals (job posting requirements), suggesting genuine preference shifts. Third, cross-country synchronicity despite varied institutional contexts points toward common technological or organizational drivers rather than country-specific policy changes.

Organizational and Individual Consequences of Early-Career Hiring Declines

Organizational Performance and Long-Term Capability Impacts

Reduced early-career hiring creates distinct organizational challenges beyond immediate staffing needs. Junior workers serve multiple functions: they provide labor at tasks where experience matters less, they offer fresh perspectives that challenge organizational inertia, and they constitute the talent pipeline for future leadership. Organizations that curtail junior recruitment may achieve short-term cost savings while eroding long-term adaptive capacity.

The human capital accumulation literature emphasizes that both firm-specific and general skills develop primarily through early-career work experiences (Arrow, 1962; Mincer, 1974). When organizations systematically reduce entry pathways, they potentially create coordination failures—all firms benefit from a skilled workforce, but individual firms face incentives to hire rather than develop talent if training investments cannot be fully captured.

Organizational knowledge research suggests that institutional memory and capability reside partly in junior-senior interaction patterns. Experienced workers transfer tacit knowledge to novices through observation, mentorship, and collaborative problem-solving. While automated tools may substitute for some knowledge transfer, questions remain about whether they can fully replicate the contextual, relationship-embedded nature of organizational learning.

Individual Career Development and Labor Market Scarring

For affected individuals, missed early-career opportunities may impose significant long-run costs. Research on recession cohorts shows that delayed entry or reduced initial job quality predicts lower lifetime earnings, slower career progression, and diminished satisfaction extending across decades (Pallais, 2014; Oreopoulos et al., 2012; Schwandt and von Wachter, 2019). These "scarring effects" operate through multiple channels: foregone skill development during critical periods, weaker professional networks, and signaling dynamics that amplify initial disadvantage.

The current contraction may prove particularly consequential if concentrated in occupations traditionally offering strong returns to early-career investment—professional services, technology, finance, and creative industries. Workers locked out of these pathways face difficult choices: accept positions in less-preferred sectors, delay labor force entry for additional education, or compete for declining entry opportunities with increasingly crowded applicant pools.

Distributional concerns also arise. If AI and remote work jointly reshape junior hiring, the effects may fall unevenly across demographic groups based on differential access to networks, credentials, or geographic flexibility that help workers bypass formal entry barriers.

Evidence-Based Organizational Responses

Table 1: Organizational Strategies for Early-Career Talent in the AI and Remote Era

Organization	Adaptation Category	Specific Intervention or Program	Implementation Details	Key Objective	Evidence or Rationale
Automattic	Structured Onboarding	Intensive Multi-week In-person Onboarding	New hires attend in-person company meetups for several weeks before transitioning to fully remote work.	Establishing relationships and cultural understanding to support subsequent distributed collaboration.	Research by Aksoy et al. (2025) shows that initial in-person onboarding improves retention and performance for remote workers.
GitLab	Structured Onboarding	Remote-first Documentation and Mentorship	Creation of explicit guides, recorded training sessions, and structured asynchronous mentorship pairings.	Reducing reliance on tacit knowledge transfer and facilitating learning across time zones.	Deliberate intervention in communication structures mitigates the tendency of remote work to create siloed networks.
Boston Consulting Group	AI Augmentation	Restructured Junior Consultant Roles	AI handles data preparation/synthesis while junior staff are trained in prompt engineering and output validation.	Emphasizing higher-value analysis and client interaction over manual data tasks.	Noy and Zhang (2023) find larger relative productivity gains from AI for less-experienced workers.
Bloomberg	AI Augmentation	AI-Assisted Analyst Roles	Junior professionals learn to quality-check and contextualize AI-generated financial research outputs.	Building AI literacy as a core competency for early-career financial professionals.	AI can reduce entry barriers by augmenting junior capabilities when workflows are redesigned.
Duolingo	AI Augmentation	Redefined Content Creation Roles	Roles shifted to include quality assurance, pedagogical design, and localization judgment of AI-generated drafts.	Expanding teams by focusing human expertise on refinement rather than initial generation.	AI adoption correlates with changing task composition rather than outright job elimination.
Shopify	AI Augmentation	AI-First Strategic Shift & Reskilling	Publicly detailed changes in job requirements and offered reskilling resources while maintaining hiring targets.	Mitigating uncertainty and supporting adaptation through procedural justice and clear roadmaps.	Transparent communication predicts employee acceptance and reduces turnover during transformation.
Deloitte UK	Hybrid Collaboration Models	Anchor Weeks and Assigned Mentorship	Periodic in-person intensive work sessions combined with senior mentors responsible for regular video-based feedback.	Reducing learning friction and ensuring regular feedback for junior consultants.	Emanuel et al. (2026) find that minimal in-person contact (one day monthly) enhances feedback quality and productivity.
Microsoft	Hybrid Collaboration Models	Collaboration Hours	Teams work synchronously (often in-person) for intensive interaction while preserving flexibility for solo tasks.	Optimizing location based on task type to support junior-senior interaction.	Proximity to senior colleagues particularly benefits junior workers (Emanuel et al., 2024).
HSBC	Hybrid Collaboration Models	Learning Neighborhoods	Office footprints redesigned specifically for junior-senior collaboration and mentorship rather than individual desks.	Optimizing scarce in-office time for high-value interaction and training.	Strategic hybrid models mitigate learning and supervision costs that disproportionately affect early-career workers.
Salesforce	Hybrid Collaboration Models	Mentor Pairing Expectations	Senior employees commit to regular in-person or high-bandwidth virtual sessions with junior colleagues.	Ensuring organic mentor networks are replaced by assigned relationship accountability.	Assigned relationship accountability is more effective for remote junior development than assuming organic networks will form.
Unilever	Hybrid Collaboration Models	Output-Based Performance Evaluation	Redesigned evaluation for remote workers focusing on impact metrics rather than physical presence.	Clarifying expectations for junior staff working outside traditional supervision models.	Remote work creates supervision friction that requires deliberate management shifts to output-based metrics.
IBM	Alternative Pathways	Skills-Based Hiring (No-Degree Roles)	Eliminated four-year degree requirements, prioritizing portfolios and technical assessments.	Creating pathways for non-traditional candidates while maintaining quality standards.	Firms value demonstrated capability over formal experience when it is reliably observable.
Google	Alternative Pathways	Residency and Apprenticeship Programs	Structured, time-limited roles with explicit learning objectives for early-career talent.	Treating early-career hiring as an investment in long-term capability rather than immediate productivity.	Portfolio-based assessments identify capable workers when traditional entry-level tasks are automated.
EY (Ernst & Young)	Alternative Pathways	Apprenticeship Programs	Combined work experience with formal qualification as a direct alternative to graduate recruitment.	Bypassing conventional progression ladders that may be economically unviable.	Apprenticeships effectively identify capable workers through demonstrated rigor.
Siemens	Alternative Pathways	Future Skills Frameworks	Identifies capabilities that retain value against automation to guide junior employee development.	Reducing anxiety about skill obsolescence and directing development efforts effectively.	Continuous skill development infrastructure treats learning as integral to work in flux environments.

Organizations confronting early-career hiring challenges can draw on both established management research and emerging evidence about AI integration and remote work. While we must be cautious about overstating what the evidence definitively shows, several intervention categories appear promising.

Structured Virtual Onboarding and Development Programs

Remote and hybrid work environments appear to create supervision and learning friction, but emerging evidence suggests these challenges need not be insurmountable. Lambert and Schindler cite research showing that even minimal in-person contact can enhance remote work effectiveness for junior employees (Aksoy et al., 2025; Aksoy et al., 2026; Emanuel et al., 2026).

Organizations experimenting with structured approaches to remote junior development include:

Automattic (WordPress parent company) operates fully remotely but requires new hires to complete intensive multi-week in-person onboarding at company meetups. This approach appears designed to establish relationships and cultural understanding that support subsequent distributed collaboration—though rigorous evidence on its effectiveness compared to alternatives is limited.

GitLab has adopted "remote-first" documentation and communication protocols that may reduce reliance on tacit knowledge transfer. Their publicly documented approach emphasizes explicit guides, recorded training sessions, and structured mentorship pairings that function asynchronously across time zones. Whether this fully compensates for in-person learning opportunities remains an open empirical question.

Deloitte UK reportedly redesigned its graduate consulting program around "anchor weeks"—periodic in-person intensive work sessions that bookend remote project work—combined with assigned senior mentors responsible for regular video-based feedback. The long-term outcomes of this model relative to traditional approaches have not been publicly evaluated.

Promising practices appear to share common elements: deliberately structured interaction replacing ad-hoc proximity-based learning, explicit documentation of tacit organizational knowledge, assigned relationship accountability rather than assuming organic mentor networks will form, and recognition that remote junior development requires more intentional design than in-person equivalents. However, we should be cautious about declaring any particular approach definitively "effective" without rigorous comparative evidence.

AI-Augmented Rather Than AI-Replacement Models

If AI tools reduce demand for certain junior tasks, organizations face a choice: eliminate those roles entirely or restructure them around complementary activities where human judgment, creativity, or contextual understanding matter most.

Some researchers emphasize AI's potential to reduce entry barriers by augmenting junior capabilities rather than replacing junior workers entirely (Brynjolfsson et al., 2025b; Althoff and Reichardt, 2026). Research on AI-assisted writing found productivity gains, with some evidence suggesting larger relative gains for less-experienced workers (Noy and Zhang, 2023). However, whether these experimental findings translate to sustained employment effects remains uncertain.

Organizations pursuing what they describe as augmentation strategies include:

Boston Consulting Group has publicly discussed restructuring junior consultant roles to emphasize higher-value analysis and client interaction, using AI tools to handle data preparation while training early-career staff on prompt engineering, output validation, and insight communication. Whether this maintains equivalent junior hiring volumes to pre-AI periods is unclear from public information.

Duolingo expanded certain content creation functions despite AI language model adoption, reportedly redefining roles to include quality assurance, pedagogical design, and localization judgment—activities that may benefit from AI draft generation while requiring human expertise for refinement. Long-term employment trends are difficult to assess from outside the organization.

Bloomberg has described creating hybrid analyst roles where junior professionals learn both traditional analysis and how to quality-check and contextualize AI-generated outputs, treating AI literacy as a core competency. Whether this approach maintains, reduces, or transforms junior hiring is an empirical question requiring data not publicly available.

These examples suggest that junior displacement may not be technologically determined but rather reflects organizational choices about task allocation and workflow design. However, we should distinguish between organizational descriptions of their strategies and rigorous evidence of their labor market impacts.

Transparent Communication and Procedural Justice

Both remote work friction and AI-driven task restructuring create uncertainty for junior workers about role expectations, career trajectories, and skill investment priorities. Organizational research on change management suggests that clear communication about transformation processes predicts better employee outcomes, though the specific mechanisms remain debated.

Organizations emphasizing transparency in their AI and remote work transitions include:

Shopify publicly communicated its "AI-first" strategic shift while detailing how this changed job requirements and career paths. The company committed to maintaining entry-level hiring targets even as role content evolved, though verifying this commitment against actual hiring data would require access to internal metrics.

Siemens has implemented "future skills" frameworks that identify capabilities likely to retain value as automation advances. This approach aims to help junior employees direct development efforts, though whether it reduces anxiety or improves outcomes is difficult to assess without employee survey data.

Unilever redesigned performance evaluation for remote/hybrid workers to emphasize output and impact metrics rather than activity indicators. This aims to clarify expectations for junior staff working outside traditional supervision models, though the effectiveness of this shift likely depends heavily on implementation details.

Effective communication practices likely involve acknowledging uncertainty rather than overpromising stability, providing resources for adaptation rather than merely announcing change, and maintaining institutional commitment to workforce development even as its mechanisms evolve. However, separating genuine commitment from public relations messaging requires insider access rarely available to external researchers.

Differentiated Talent Development Pathways

If traditional entry-level task bundles become economically unviable—whether due to AI, remote friction, or other forces—organizations might develop alternative qualification mechanisms that bypass conventional progression ladders.

Organizations experimenting with alternative entry pathways include:

IBM eliminated four-year degree requirements for many technical roles, instead emphasizing demonstrated skills via coding portfolios, open-source contributions, and technical assessments. This creates pathways for non-traditional candidates while potentially maintaining quality standards, though the long-term career outcomes of workers entering through different pathways is an important empirical question.

EY (Ernst & Young) launched apprenticeship programs in multiple markets, combining work experience with formal qualification. These are framed as alternatives to traditional graduate recruitment rather than lower-status options, though whether they are perceived and function equivalently is unclear.

Google has expanded residency and apprenticeship programs that provide structured, time-limited junior roles with explicit learning objectives. These programs appear to treat early-career hiring as an investment in long-term capability, though the proportion of total junior hiring they represent is not publicly disclosed.

These models acknowledge that if traditional junior task bundles change significantly, organizations may need alternative mechanisms for assessing potential and enabling skill development. However, whether these alternatives prove equally effective at human capital formation is a question requiring longitudinal data.

Hybrid Operating Models with Deliberate Collaboration Structures

Rather than treating remote work as binary (fully in-office vs. fully remote), organizations can design hybrid models that optimize around specific learning and supervision needs. Lambert and Schindler cite research suggesting structured hybrid arrangements can preserve benefits while mitigating certain costs (Bloom et al., 2015; Emanuel et al., 2024).

Organizations implementing learning-optimized hybrid models include:

Microsoft has described adopting "collaboration hours" when team members work synchronously (often in-person) for activities requiring intensive interaction, while preserving flexibility for focused individual work. This acknowledges that different tasks may have different location optima, though the optimal balance likely varies across roles and individuals.

Salesforce created mentor pairing expectations where senior employees commit to regular in-person or high-bandwidth virtual sessions with assigned junior colleagues, combined with flexibility for other work modes. Effectiveness likely depends heavily on whether these commitments are genuinely enforced and valued in performance evaluation.

HSBC restructured office footprints around "learning neighborhoods"—spaces designed for junior-senior collaboration—rather than individual desks, optimizing scarce in-office time for high-value interaction. Whether this design philosophy translates to measurable improvements in junior development is an open question.

These approaches recognize that remote work friction may not be uniformly distributed across all activities. Strategic hybrid models might preserve flexibility benefits while mitigating costs that disproportionately affect early-career workers. However, the "optimal" hybrid model likely varies substantially across organizational contexts, making generalizable recommendations difficult.

Building Long-Term Organizational Capability in an Uncertain Environment

Beyond immediate interventions, organizations must develop systemic capabilities for managing workforce composition amid ongoing technological and organizational change.

Continuous Learning Systems and Adaptive Skill Portfolios

Rather than assuming static skill requirements, organizations increasingly need mechanisms for ongoing capability assessment and development that adapt as roles evolve. This likely involves treating skill development as integral to work rather than occasional, emphasizing portfolios over rigid job titles, and implementing internal mobility systems that redeploy workers as organizational needs shift rather than defaulting to external hiring and layoffs.

Technology companies have pioneered some of these practices—internal talent marketplaces, rotation programs, structured learning time—but applicability to other sectors is uncertain. The core insight is that rigid job architectures designed for stable environments may become liabilities when both technology and organizational structure are in flux. However, more flexible approaches also create challenges around career progression clarity and performance evaluation.

Distributed Leadership and Voice Mechanisms

Decisions about AI adoption, work location policies, and hiring strategies have profound consequences for workers yet are often made through narrow optimization of short-term costs or executive preferences. Organizations building long-term capability might create mechanisms for broader input, though the forms this takes vary widely across contexts.

Some European firms have formal works councils that negotiate technology adoption and work arrangement policies. Other organizations use employee resource groups, regular surveys, or structured feedback sessions. A few experiments involve worker representatives in technology evaluation committees.

The underlying principle is that those affected by organizational change often possess valuable information about implementation challenges, unintended consequences, and alternative approaches. Mechanisms that surface this knowledge may improve decision quality while building legitimacy. However, such mechanisms can also slow decision-making and create political complexity, suggesting tradeoffs rather than unambiguous benefits.

Data-Informed Workforce Planning with Ethical Guardrails

Organizations often make hiring and restructuring decisions based on limited evidence about actual productivity impacts of different workforce compositions, technologies, or work arrangements. Building better feedback loops could improve decision quality, though this raises significant privacy and equity concerns.

Promising approaches might include workforce analytics that measure capability deployment and skill gaps, structured experiments that randomize work arrangements where feasible, and early warning systems that detect emerging patterns. However, such systems require protecting worker privacy, avoiding discriminatory proxies, and maintaining transparency about what is measured and why. The goal should be organizational learning, not invasive surveillance—a line that proves difficult to maintain in practice.

Methodological Considerations and Interpretive Caution

The Lambert and Schindler (2026) study exemplifies both the promise and perils of exposure-based research designs for understanding technological change.

The Correlated Treatment Problem

The study's central finding—that WFH exposure remains predictive while GenAI exposure attenuates in joint specifications—is methodologically interesting but interpretively ambiguous. High correlation between exposures (Spearman rank ρ ≈ 0.77 at the occupation level) creates fundamental challenges for separately identifying effects even with difference-in-differences designs.

The authors employ two occupation-level exposure indices: the WFH measure from Hansen et al. (2023) and the GenAI exposure index from Eloundou et al. (2024). When entered separately in regression models, both predict declines in junior hiring. When entered jointly, WFH coefficients remain while GenAI coefficients attenuate toward zero.

Standard econometric approaches assume that conditioning on observed factors eliminates confounding. But when two treatments are highly correlated and potentially measured with error, joint specifications can arbitrarily attribute shared effects to whichever treatment is measured more precisely, enters the model in particular functional forms, or happens to have slightly lower correlation with unobserved confounders.

The authors conduct extensive robustness checks—alternative measures, non-parametric controls, measurement error simulations—and these generally support their WFH-dominant interpretation. However, these cannot fully resolve the fundamental challenge that WFH and GenAI may operate through partially overlapping mechanisms affecting similar occupations via similar pathways.

From Exposure to Actual Adoption

Occupation-level exposure indices—whether for remote work feasibility or AI task overlap—measure technological potential rather than actual implementation. The gap between potential and practice is substantial and varies systematically across organizations, regions, and time periods.

The authors partly address this by examining actual WFH adoption measured through job posting language (whether postings explicitly mention remote/hybrid options). They find that actual WFH adoption in 2021-2022 predicts subsequent junior hiring declines in 2023-2025, providing some validation that their exposure measures track real organizational behavior.

However, this approach introduces different challenges: selection (which firms advertise remote options may differ systematically from those that don't), measurement (posting language imperfectly captures true arrangements), and absence of comparable GenAI measures (there is no similar direct measure of GenAI adoption at the firm level with equivalent coverage).

Ideally, researchers would observe firm-specific AI and remote work implementation intensity rather than occupation-level potential, but such data rarely exists at scale. The exposure-based approach is methodologically pragmatic but introduces known limitations.

External Validity and Mechanistic Uncertainty

Even if the reported associations are causally identified within the studied sample, generalization requires caution. The data sources (resume-derived hiring records, online job postings) may not represent all hiring activity—they likely oversample larger firms, white-collar occupations, and workers with online professional presence. The time period (through 2025) is relatively early in both generative AI and post-pandemic remote work maturity. The studied countries share institutional similarities that may not extend to other regions.

More fundamentally, statistical associations—even well-identified ones—do not fully reveal mechanisms. Why does remote work predict reduced junior hiring? Is it supervision costs? Learning friction? Changes in organizational culture? Selection of firms into remote work based on other characteristics that independently affect hiring?

Similarly, if GenAI effects are smaller than remote work effects in these specifications, we face multiple possible interpretations:

• AI genuinely has limited impact on junior tasks (contradicting the technological substitution hypothesis)

• Adoption remains nascent and effects will emerge later (timing)

• Firms are successfully augmenting junior workers rather than replacing them (adaptation)

• The exposure measures capture AI potential poorly (measurement error)

• High correlation with WFH makes separate identification essentially impossible (statistical confounding)

The authors favor the interpretation that WFH is the dominant driver, but the evidence permits alternative readings. Policy and practice recommendations require mechanistic understanding beyond predictive coefficients.

Sample Selection and Coverage Considerations

The resume-based hiring data likely systematically underrepresents certain worker populations: those without online professional profiles, workers in occupations with low resume-updating rates, and potentially younger or disadvantaged workers less likely to use platforms like LinkedIn. If these groups experienced different impacts from AI or remote work, the results might not generalize.

The job posting data covers online vacancies, which may not represent all recruitment channels. Informal hiring, internal promotions, and positions filled through networks rather than public postings could follow different patterns. If AI or remote work affects different recruitment channels differently, focusing only on online postings could mislead.

These are not criticisms unique to this study—they represent general challenges in labor market research—but they counsel against treating the findings as definitive.

Toward a More Complete Understanding

Rather than asking "Is it AI or remote work?", more productive framings acknowledge multicausality, heterogeneity, and uncertainty.

Both Mechanisms Likely Operate Simultaneously

AI and remote work likely affect junior hiring through distinct causal pathways that can coexist:

• AI may genuinely reduce demand for certain entry-level analytical tasks (report generation, data cleaning, basic research) while potentially creating demand for new skills (prompt engineering, output validation, AI-assisted analysis). The net employment effect depends on magnitudes and substitution rates not yet well-established.

• Remote work may genuinely increase supervision costs and slow tacit learning, particularly for workers without established internal networks, while offering flexibility benefits that some junior workers value highly. The net effect on junior hiring likely depends on organizational capabilities for managing remote teams.

• Joint effects may arise because AI tools partially compensate for remote work friction (e.g., AI-generated documentation reduces reliance on in-person knowledge transfer) or because organizations simultaneously adopting both technologies restructure work in ways neither technology individually requires.

Disentangling these channels requires research designs that go beyond exposure-based correlations to examine actual organizational practices, worker experiences, and performance outcomes under different combinations of technology adoption and work arrangements.

Heterogeneity Likely Exceeds Average Effects

Average effects obscure important variation. Some organizations likely successfully maintain robust junior hiring while adopting AI and remote work—existence proofs that displacement is not technologically determined. Others may reduce junior hiring for reasons largely unrelated to either technology (cost pressures, changing business models, shifting skill requirements).

Understanding this heterogeneity matters because it reveals possibility and identifies potentially effective practices. Rather than debating whether "on average" AI or remote work matters more, practitioners might benefit from identifying which organizational capabilities, workflow designs, and management practices allow firms to adopt new technologies and work arrangements while preserving early-career pathways.

The Lambert and Schindler paper focuses on average treatment effects across large samples. Complementary research examining organizational variation in responses would provide valuable practical insight, though it would likely require different data and methods (firm case studies, surveys, detailed organizational records).

Longer Time Horizons and Dynamic Adjustments

Early evidence on transformative technologies often misleads about long-run equilibria. The initial impact of electricity, computers, or the internet on productivity and employment bore little resemblance to patterns observed decades later. Organizations require time to restructure work processes, workers need time to acquire complementary skills, and institutions must adapt.

Current evidence on AI and remote work reflects a narrow window when both are relatively novel. Junior hiring patterns in 2025 may poorly predict 2030 or 2035 outcomes as organizations learn to manage remote junior workers more effectively, as AI capabilities mature and adoption deepens, as educational institutions adapt curricula, as labor markets reprice junior versus senior workers, and as new entry pathways and credentialing mechanisms emerge.

Recognizing this uncertainty should temper strong causal claims while motivating continued monitoring and research. The Lambert and Schindler findings represent an important early signal but not a definitive answer.

The Limits of Statistical Decomposition

The question "How much is AI versus how much is remote work?" may be inherently unanswerable with available methods when the two forces are highly correlated, causally interrelated, and potentially operating through overlapping mechanisms.

Statistical techniques can partition variance in the data, but this partition depends on functional form assumptions, measurement choices, and the specific counterfactuals implied by the research design. Different defensible specifications can yield different answers without any being clearly "wrong."

This is not a failure of statistics but a reflection of the problem's structure. When two forces arrive simultaneously, affect the same populations through potentially related mechanisms, and cannot be experimentally manipulated, definitive causal decomposition may exceed what observational data can support.

Rather than seeking a single "correct" variance decomposition, the research community might be better served by triangulating across multiple methods—observational studies with different identification strategies, surveys asking organizations and workers directly about decision drivers, detailed case studies of organizational responses, and eventually, as time permits, longer panel data that can better separate timing and dynamics.

Conclusion

The decline in early-career hiring since 2022 represents a genuine and consequential labor market shift worthy of serious attention from practitioners, policymakers, and researchers. Both generative AI and persistent remote work plausibly contribute, operating through distinct mechanisms with potentially interactive effects.

The Lambert and Schindler (2026) working paper provides valuable evidence that remote work exposure predicts hiring changes even after controlling for AI exposure, challenging simplistic AI-replacement narratives that dominated early discourse. Their extensive robustness checks strengthen confidence in this pattern. However, this should not be read as definitively establishing remote work as "the" primary driver or as exonerating AI from causal responsibility.

Methodological realities counsel interpretive humility:

• High correlation between exposures (ρ ≈ 0.77) makes clean separation fundamentally difficult

• Measurement challenges affect both exposure indices and outcome variables

• Limited time horizons (data through 2025) may miss longer-run adjustments

• The gap between technological potential and organizational implementation complicates inference from exposure-based designs

• Sample coverage limitations in resume and job posting data may miss important populations

• Mechanistic uncertainty means we don't fully understand why these patterns emerge

For practitioners, the appropriate response is not to await definitive academic resolution but to recognize both challenges and pursue evidence-informed adaptations:

• Invest in structured approaches to remote/hybrid work that preserve learning and supervision effectiveness for early-career workers, recognizing this requires deliberate design

• Design AI integration strategies that emphasize augmentation and skill development rather than assuming displacement is inevitable

• Maintain institutional commitment to workforce development even as entry pathways necessarily evolve

• Build organizational capabilities for continuous adaptation amid ongoing change

• Use internal data and experimentation to understand what works in your specific context

• Communicate transparently with workers about changes and uncertainties

For researchers, this episode illustrates both challenges and opportunities in studying contemporaneous technological change. Exposure-based designs offer valuable early signals but require methodological care and interpretive caution. The Lambert and Schindler paper exemplifies rigorous execution of this approach while also revealing its inherent limitations.

Complementary research approaches can help triangulate:

• Firm case studies examining organizational decision-making processes

• Worker and manager surveys asking directly about factors influencing hiring decisions

• Longitudinal panel data enabling better separation of timing and mechanisms

• Direct measurement of AI and remote work adoption intensity (not just exposure)

• Experimental or quasi-experimental designs exploiting policy variation or firm-level shocks

For policymakers, the evidence base remains too uncertain to support heavy-handed interventions targeting specific technologies. However, the data clearly indicate that early-career pathways face genuine challenges requiring attention. Policies might usefully emphasize:

• Supporting organizational experimentation and knowledge-sharing around effective practices

• Ensuring educational institutions adapt to teach capabilities complementary to emerging technologies

• Monitoring labor market trends for distributional impacts and scarring effects

• Maintaining apprenticeship programs and alternative credentialing pathways

• Avoiding premature conclusions that could lead to misguided regulation of technologies with genuine benefits

The central message is one of appropriate uncertainty combined with active learning. We do not yet know with confidence whether AI, remote work, both equally, or other factors primarily drive the early-career hiring decline. The Lambert and Schindler evidence suggests remote work plays a significant role—more significant than early AI-centric narratives acknowledged. But it does not definitively close the question.

The future of early-career work will be shaped not by any single technology but by how organizations, workers, educators, and institutions collectively respond to combined challenges and opportunities. Acknowledging complexity and uncertainty, while pursuing evidence-informed adaptation and continued research, offers the most promising path forward.

Rather than declaring victory for one explanation over another, we should view the current evidence as highlighting important questions requiring deeper investigation: What organizational practices enable successful remote junior development? How do AI tools actually get used in early-career roles? Which workers and firms are most affected? What alternative pathways might emerge? These questions matter more than definitive causal attribution to one technology or another.

Research Infographic

The Early Career Paradox Slide Deck

References

1. Aksoy, C. G., Bloom, N., Davis, S. J., Marino, V., & Özgüzel, C. (2025). Remote work, employee mix, and performance. NBER Working Paper w33851.

2. Aksoy, C. G., Bloom, N., Davis, S., Marino, V., & Özgüzel, C. (2026). In-person contact improves remote-work performance. [Unpublished working paper]

3. Althoff, L., & Reichardt, H. (2026). Task-specific technical change and comparative advantage. CESifo Working Paper 12403.

4. Arrow, K. J. (1962). The economic implications of learning by doing. The Review of Economic Studies, 29(3), 155–173.

5. Azar, J., Gine, M., & Sanz-Espín, J. (2025). AI is already eroding wages: Quasi-experimental evidence from occupational exposure. SSRN Working Paper 5842084.

6. Bloom, N., Liang, J., Roberts, J., & Ying, Z. J. (2015). Does working from home work? Evidence from a Chinese experiment. The Quarterly Journal of Economics, 130(1), 165–218.

7. Brynjolfsson, E., Chandar, B., & Chen, R. (2025a). Canaries in the coal mine? Six facts about the recent employment effects of artificial intelligence. Stanford Digital Economy Lab Working Paper.

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

9. Eloundou, T., Manning, S., Mishkin, P., & Rock, D. (2024). GPTs are GPTs: Labor market impact potential of LLMs. Science, 384(6702), 1306–1308.

10. Emanuel, N., Harrington, E., & Pallais, A. (2024). Research: How remote work impacts women at different stages of their careers. Harvard Business Review.

11. Emanuel, N., Harrington, E., & Pallais, A. (2026). The power of proximity to coworkers. The Quarterly Journal of Economics, qjag027.

12. Hansen, S., Lambert, P. J., Bloom, N., Davis, S. J., Sadun, R., & Taska, B. (2023). Remote work across jobs, companies, and space. NBER Working Paper w31007.

13. Hosseini Maasoum, S. M., & Lichtinger, G. (2025). Generative AI as seniority-biased technological change: Evidence from U.S. résumé and job posting data. SSRN Working Paper 5425555.

14. Lambert, P. J., & Schindler, Y. (2026). The broken ladder: AI, remote work, and early-career hiring. [Working paper, May 2026]

15. Mincer, J. (1974). Schooling, experience, and earnings. National Bureau of Economic Research.

16. Noy, S., & Zhang, W. (2023). Experimental evidence on the productivity effects of generative artificial intelligence. Science, 381(6654), 187–192.

17. Oreopoulos, P., von Wachter, T., & Heisz, A. (2012). The short- and long-term career effects of graduating in a recession. American Economic Journal: Applied Economics, 4(1), 1–29.

18. Pallais, A. (2014). Inefficient hiring in entry-level labor markets. American Economic Review, 104(11), 3565–3599.

19. Schwandt, H., & von Wachter, T. (2019). Unlucky cohorts: Estimating the long-term effects of entering the labor market in a recession in large cross-sectional data sets. Journal of Labor Economics, 37(S1), S161–S198.

20. Teeselink, B. K. (2025). Generative AI and labor market outcomes: Evidence from the United Kingdom. SSRN Working Paper 5516798.

21. Yang, L., Holtz, D., Jaffe, S., Suri, S., Sinha, S., Weston, J., Joyce, C., Shah, N., Sherman, K., Hecht, B., & Teevan, J. (2022). The effects of remote work on collaboration among information workers. Nature Human Behaviour, 6(1), 43–54.

---

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). Organizational AI Transparency and Employee Resilience: Building Trust, Autonomy, and Confidence in Hybrid Work. Human Capital Leadership Review, 35(1). doi.org/10.70175/hclreview.2020.35.1.2