Research Note: Beyond Adoption: Intensity and Integration as the Missing Link in Yotzov et al.'s 'Firm Data on AI'

Yotzov et al. (2026), in their NBER working paper ‘Firm Data on AI’ provide rare, international survey evidence on firm AI adoption and perceived impacts. The headline result, that most firms are reporting little realized productivity or employment impact so far, will potentially be used by skeptics as proof that AI is hype. That reading is too blunt. The paper’s own measures suggest that, for many firms, “adoption” still means low-intensity, weakly integrated use, which is exactly the condition under which large economic effects should be minimal. The correct conclusion is not that AI lacks value, but that organizational capability, governance, and workflow integration remain the binding constraints. This note argues that the binding constraint is not model capability but organizational integration, and it proposes AI-OS as a disciplined operating method for converting diffuse AI use into governed, measurable changes in work.