Organizational Technology Ladders: Remote Work and Generative AI Adoption
A · Strong evidence Single-author arXiv paper by Gregor Schubert (UCLA Anderson) using near-universe US job postings and two instrumental-variable strategies, with robustness checks and an earnings-call validation; clear method, checkable estimates.
Summary based on the full text·Original: about 75 minutes to read
In short
Gregor Schubert (UCLA Anderson, 13 August 2026) argues that firms climb an "organisational technology ladder": adopting one technology changes hiring, infrastructure and skills in ways that make the next one cheaper to adopt. Firms pushed into more remote hiring during 2021-22 later adopted generative AI faster. A 10-point higher remote hiring share raises the share of job postings mentioning generative AI in 2023-24 by 0.4 points across firms and 0.7 points across occupations within firms. Remote work shifted hiring toward technical and managerial skills that convert AI exposure into adoption. Firms that issued return-to-office mandates show a larger AI response, which the author reads as using automation to escape the productivity cost of distributed teams.
Key numbers
- 0.4 percentage pointsrise in the share of postings mentioning generative AI in 2023-24 per 10-point increase in a firm's remote hiring in 2021-22
- 0.7 percentage pointssame effect measured across occupations within firms
- Jan. 2010 to Sep. 2024span of the Lightcast online job-posting data used
As reported by the source.
English·Web page·opens arxiv.org
What it reports
- Causal estimates: +0.4 pp generative AI posting share per +10 pp remote hiring across firms; +0.7 pp within firms across occupations.
- Effects are strongest in information- and decision-intensive sectors such as finance and business services.
- Excluding the tech sector halves the across-firm estimate but leaves the within-firm estimate almost unchanged, so the effect is not confined to tech firms.
- Remote work adoption raised hiring for roles needing decision-making, social and managerial skills, for positions requiring five or more years of experience, and for technology skills.
- Ladder effects are larger for firms that already hired for tech skills, complex roles and managerial roles.
- Earnings calls (Capital IQ, classified with an LLM pipeline) link remote work to data, cloud, remote-access and security investments that firms later cite as enablers of generative AI.
- A task-based model formalises remote work as raising worker productivity at the cost of decision quality, and AI as automation whose cost falls with prior IT investment.
How they got it
Lightcast near-universe US online job postings (staffing firms, internships and part-time jobs removed). Generative AI adoption is measured by Lightcast skill tags in posting text, validated against earnings-call mentions. Main instrument: a firm's exposure to labour markets where competitors are likely to offer remote work; alternative instrument: pre-pandemic commuting times. Within-firm estimates interact these with occupation teleworkability and include firm, occupation and metro fixed effects.
Limits
- Postings measure stated hiring demand and formal practice, not actual AI use by staff.
- Instrument validity depends on labour-market pressure affecting AI only through remote work, despite controls.
- The return-to-office interpretation is inferential.
- Single-author working paper, US only, data to September 2024.
Why it matters for companies
Adoption is path-dependent: the digital infrastructure, data access and management skills built for distributed work are the same ones AI deployment needs. Firms that treated remote work as a temporary concession and invested little are likely to find AI adoption slower and more expensive. For workforce strategy, this argues for counting prior digital capability, not AI tool spend, as the predictor of how fast a business unit will absorb AI, and for investing in managers who can run work through systems rather than proximity.
Summary based on the full text. Read the original for the full argument and data.Read the original on arxiv.org
- Tags
- AI adoption