AI & Work
Buying AI tools and building a productive AI-enabled organization are not the same thing. An August NBER working paper by Tania Babina, Alex X. He and Renhao Jiang of the University of Maryland argues that recent productivity gains from artificial intelligence are tied to the organizational knowledge companies build as they hire AI-skilled employees and learn to use the technology—what economists call organization capital.
How the study measures AI investment
The researchers measured firms’ AI investment by their employment of AI-skilled workers, a category they say spans “machine learning through generative and agentic AI.” They combined Lightcast data covering nearly all U.S. online job postings with Revelio Labs résumé data covering about two-thirds of the U.S. workforce, linked the result to U.S. public companies, and built a separate measure of jobs that create organization capital from workers’ job descriptions.
By 2024, they report, U.S. public firms employed more than 300,000 AI-skilled workers, 2.1% of their U.S. workforce. Over 2018 to 2024, a one-standard-deviation increase in the growth of a firm’s AI worker share was associated with 6.2 percentage points greater growth in sales per worker and 7.5 points greater growth in revenue total factor productivity over six years, “roughly one percentage point of additional productivity growth per year.” The authors found no comparable relationship over 2010 to 2018.
What organization capital means
The paper defines organization capital as “durable firm-specific knowledge acquired through learning-by-doing that enables more efficient production.” Its examples include Morgan Stanley training an AI assistant on its own advisors’ meeting notes and Expedia turning engineers’ incident histories into agentic workflows. None of that appears as a line item in a software subscription. The authors find that AI jobs are almost twice as likely as non-AI jobs to build organization capital, 58% versus 31%, and that 66% of generative and agentic AI jobs do so, compared with 57% of machine-learning, natural-language-processing and computer-vision jobs.
Productivity is not headcount
The findings are distinct from claims that AI necessarily eliminates jobs, but they are not reassuring about hiring volume either. Over 2018 to 2024, the authors write, “AI investment predicts substantial sales growth but no employment growth.” AI-investing firms shifted toward workers with advanced degrees and senior roles, while the shares of non-college and junior workers declined. A company can invest in AI skills while changing the composition of its workforce more than its size.
The study uses observed firm data, and an association between AI-skilled employment and productivity is not a guaranteed return on any particular new hire. The authors do add an instrumental-variable test, based on firms’ pre-2010 hiring ties to universities strong in AI research, which they say supports a causal interpretation. NBER working papers have not been peer-reviewed.
What executives can track
For business readers, the operational implication is that staff capability matters alongside software spending. When evaluating an internal pilot, executives can track learning as well as task time. Does a team develop reusable procedures? Can new employees adopt them? Do quality and delivery hold up after the initial enthusiasm fades? Those are tests of whether an investment has become organizational capability. The paper provides a framework for asking them, not a universal implementation recipe or a forecast of employment outcomes.
Sources
- Tania Babina, Alex X. He and Renhao Jiang, “Canaries in the Gold Mine: Early Productivity Gains from Artificial Intelligence Creating Organization Capital,” NBER Working Paper 35684, August 2026. https://www.nber.org/papers/w35684
Featured photo: Planners post sticky notes on issue boards during a recovery-planning session. Representative image of teams building shared process knowledge. Photo: Robert Kaufmann/FEMA, via Wikimedia Commons (Public domain). Cropped from the original.
