Firms that embraced remote work during COVID were measurably more likely to formalize generative-AI hiring afterwards: a 10‑point rise in remote hiring led to roughly a 0.4–0.7 percentage-point uptick in generative‑AI job postings, suggesting prior digital and managerial investments create a ladder to faster AI adoption.

Main Finding

Firms that adopted more remote work during the COVID period were more likely to adopt generative AI soon after. Using an IV strategy, the paper estimates that a 10 percentage-point (pp) increase in a firm's remote-hiring share in 2021–2022 causally raises the share of job postings mentioning generative AI in 2023–2024 by about 0.4 pp at the firm level and by about 0.7 pp across occupations within firms. Evidence indicates this is driven by infrastructure and skill complementarities plus organizational frictions—i.e., climbing an “organizational technology ladder” makes later technology adoption easier.

Key Points

• Technology ladder concept: Adoption of one organizational technology (remote work) creates reusable investments in digital infrastructure, skills, and organizational practices that lower the cost of adopting the next technology (generative AI).
• Magnitudes (causal, IV): +10 pp remote hiring → ≈ +0.4 pp generative-AI mentions (firm-level); within-firm occupation variation: +0.7 pp.
• Mechanisms supported by evidence:
  • Infrastructure complementarity: Remote work induced investments in cloud, remote access, data infrastructure and security that enable generative AI deployment.
  • Skill accumulation and hiring shifts: Remote adoption increased hiring for technical, managerial, and decision-making roles that facilitate AI implementation.
  • Organizational frictions: Firms that mandated return-to-office (RTO) show larger generative-AI responses to remote hiring—consistent with firms using AI to mitigate coordination/productivity issues from distributed teams.
• Heterogeneity: Effects are stronger in information- and decision-intensive sectors (finance, business services) and for firms/occupations with higher pre-existing tech and managerial skill demand.
• Robustness: Results hold controlling for firm suitability for remote work/AI automation, pre-pandemic technology capability, industry fixed effects, alternative IV (commuting times), within-firm fixed effects, MSA controls, and different remote definitions.

Data & Methods

• Primary data:
  • Lightcast near-universe U.S. online job postings (Jan 2010–Sep 2024; main analyses use postings through Sep 2024).
  • Labels from Lightcast identify remote/hybrid postings and generative-AI mentions in job text.
  • Occupation exposure to generative AI: occupation-level score (Eisfeldt et al. 2025 / based on Eloundou et al. 2023).
  • Occupation characteristics from O*Net (decision-making, social skills, teleworkability, leadership proxies).
  • Employment and wage by occupation/MSA from BLS OES; commuting times from IPUMS NHGIS (ACS); RTO policies from Flex Index; earnings-call transcripts from Capital IQ used for qualitative/validation analysis.
• Outcome and timing:
  • Remote share measured over 2021–2022 (excluding Q4 2022 to avoid ChatGPT release effects).
  • Generative-AI mentions measured in job postings in 2023–2024 (up to Sep 2024).
• Identification strategy:
  • Main IV: firm exposure to labor-market pressure to offer remote work—constructed from predicted differences in local labor-market competition with firms more likely to offer remote options (pre-pandemic spatial labor-market variation).
  • Alternative IV: firm pre-pandemic exposure to labor markets with longer commuting times (proxying higher employee demand for remote).
  • Within-firm occupation IV: interaction of firm labor-market exposure with occupation teleworkability to isolate occupation-level differential remote pressure.
  • Specifications include firm, occupation, and (in some tests) MSA fixed effects; controls for pre-pandemic hiring patterns, IT capability proxies, industry fixed effects.
• Mechanism checks:
  • Changes in job-skill demand (machine learning, data, managerial/decision-making skills) measured from postings.
  • Heterogeneity by sector, pre-existing tech capabilities, and RTO mandate status.
  • Earnings-call text analysis (multi-stage LLM classifier) documents narratives linking remote-work investments (cloud, data infra, security) to generative-AI readiness.
• Validity and robustness:
  • Multiple instruments and alternative samples produce consistent estimates.
  • Within-firm occupation analyses address firm-level confounders.
  • Authors discuss channels consistent with instruments and perform controls for observable confounders; RTO heterogeneity provides suggestive evidence distinguishing channels (skill vs. organizational-friction).

Implications for AI Economics

• Organizational path dependence: Technology adoption is serially interdependent—early organizational choices (e.g., remote work) shape firms’ marginal costs and benefits for later technologies. Models and empirical work on technology diffusion should incorporate these sequential complementarities.
• Compounding firm heterogeneity: Sequential complementarity can amplify productivity divergence across firms. Firms that invested earlier in digital infrastructure and relevant skills may capture early AI productivity gains, potentially creating persistent winners and laggards.
• Labor demand and skill composition: Remote-work-driven hiring shifted demand toward technical, managerial, and decision-making skills—skills that both complement and facilitate AI adoption. Policy and training programs should account for how prior organizational technology choices reshape future skill demand.
• Role of organizational frictions: Firms facing coordination/productivity costs from remote arrangements may lean more heavily on automation to restore performance—implying that firm-level organizational constraints can shape the direction and speed of AI adoption.
• Policy and firm strategy:
  • To reduce unequal diffusion, interventions could target complementarities (expand access to cloud/data infrastructure, subsidize retraining in relevant skills).
  • Anticipating sequential complementarities can inform firm investment strategy: investing in reusable digital infrastructure and managerial capacity can raise optionality for future tech waves.
• Future research directions:
  • Directly quantify productivity and wage consequences of the technology ladder (beyond job-posting proxies).
  • Explore worker-level outcomes (wage, mobility) from being employed at firms that climbed the ladder.
  • Analyze longer-run dynamics: do early adopters sustain advantages as technologies mature, and what policies can mitigate widening gaps?

Short summary: The paper provides causal evidence that pandemic-era remote-work adoption materially increased firms’ subsequent generative-AI hiring, and traces this to reusable infrastructure investments, skill accumulation, and organizational responses—arguing for a sequential “organizational technology ladder” that amplifies firm-level divergence in technology uptake and likely productivity.