Welcome back to TPI’s Research Roundup, our semi-regular compilation of recent outside research of interest to tech policy nerds. DISCLAIMER: Not all authors are affiliated with TPI. We do not necessarily agree with everything, or even anything, in these papers, but find them interesting.
In this Issue: Four new economics papers on AI adoption among firms and its impact on labor markets, from the Federal Reserve Bank of New York, MIT, and University of Toronto.
1. AI Uptake by Industrial Firms
The Adoption of Industrial AI in America
McElheran, Kristina, Mu-Jeung Yang, Zachary Kroff, and Erik Brynjolfsson
AEA Papers and Proceedings 116: 20–25. 2026.
Their question: What is the rate of AI adoption among U.S. industrial firms, and what factors drive or limit it?
Their answer: In collaboration with the U.S. Census Bureau, survey data from manufacturing firms showed that in 2021, fewer than 23 percent of firms had adopted even one AI-related technology or application. Surprisingly, better prior performance did not predict AI uptake: establishments with higher labor productivity in 2019 were less likely to report AI use by 2021. Respondents cited cost, a lack of clear use cases, and a lack of skills to implement AI as the main barriers to adoption.
Why it matters: This survey-based snapshot reveals uneven uptake among firms. The finding that higher-performing firms are slower to adopt AI suggests that barriers like cost, skills, and proven use cases matter more than general digital readiness.
2. Productivity Impacts of AI on Firms
The Rise of Industrial AI in America: Microfoundations of the Productivity J-curve(s)
Kristina McElheran, Mu-Jeung Yang, Erik Brynjolfsson, and Zachary Kroff
2025. Preprint, SSRN
Their question: Is AI making industrial firms more productive?
Their answer: A productivity “J-curve” shows that productivity gains decline at first, then rise over the long-run after AI deployment.
Why it matters: Understanding how AI increases productivity—in what manner, and for which firms—is important for policymakers as they weigh regulation against effects on labor markets and global competitiveness. A seminar talk on this paper from May 11, 2026 at Stanford is available at https://www.youtube.com/watch?v=pKYydu-7SwM.
3. The 2025 AI Agent Index
The 2025 AI Agent Index: Documenting Technical and Safety Features of Deployed Agentic AI Systems.
Staufer, Leon, Kevin Feng, Kevin Wei, Luke Bailey, Yawen Duan, Mick Yang, A. Ozisik, Stephen Casper,
and Noam Kolt
ACM FAccT 2026
Their question: What are the origins, design, capabilities, ecosystem, and safety features of prominent AI agents?
Their answer: U.S. and Chinese agent developers appear to take markedly different approaches to safety disclosure. Of the 30 agents studied, 21 are U.S.-incorporated and five are Chinese, but Chinese developers rarely publish documented safety frameworks or compliance standards—only one of the five did so in each category. Across the full sample, only half of developers published AI safety frameworks, and enterprise assurance standards such as SOC 2 and ISO 27001 are more common than agent-specific safety disclosures.
Why it matters: As the AI agent economy expands with global competition, it’s critical to have data on the products and companies releasing these agents. The 2025 AI Agent index, released in 2026, is available at https://aiagentindex.mit.edu/.
4. AI’s Impacts on Labor Markets
Do Job Postings Show Early Labor‑Market Effects of AI?
Richard Audoly, Miles Guerin, and Giorgio Topa
Liberty Street Economics, Federal Reserve Bank of New York, May 14, 2026.
Their question: Have job postings for AI-exposed occupations declined since the release of ChatGPT in late 2022?
Their answer: Job posting data shows little evidence of a decline in labor demand for AI-exposed jobs, despite an overall slowdown in hiring since 2022.
Why it matters: If AI were destroying jobs, job postings would be one place to find evidence. It may be that the period in late 2022 is too early to determine AI’s effects, since the early models were not as powerful as today’s—or tomorrow’s. Even so, this study’s design can be applied to more recent periods to test the hypothesis again.
If you’ve read a paper you think might be interesting to include in the next Research Roundup, feel free to send it to us at [email protected].
Sarah Oh Lam is a Senior Fellow at the Technology Policy Institute. Oh completed her PhD in Economics from George Mason University, and holds a JD from GMU and a BS in Management Science and Engineering from Stanford University. She was previously the Operations and Research Director for the Information Economy Project at George Mason School of Law. She has also presented research at the 39th Telecommunications Policy Research Conference and has co-authored work published in the Northwestern Journal of Technology & Intellectual Property among other research projects. Her research interests include law and economics, regulatory analysis, and technology policy.
