AI adoption beats frontier hype: diffusion and skill infrastructure over model races

The article argues that the biggest advantage in the AI tech sector will come from AI adoption, not just from building frontier models. Drawing on political scientist Jeff Ding’s “diffusion theory,” it contrasts the “leading sector” idea (who invents the biggest new industry wins) with the diffusion view: general-purpose technologies create long-term power when they are embedded across ordinary work over decades. A key mechanism is “skill infrastructure”—education and training systems that broaden engineering capacity, standardize best practices, and link research to industry. Inside companies, the author says AI strategy often becomes either a procurement decision (choosing the best vendor/model/tool) or a moonshot lab demo. The diffusion approach shifts focus to organizational know-how: redesigning workflows, making learning shareable, and compounding improvements. The piece frames successful enterprise AI as diffusion that is organizational, not merely technical. It highlights how early electrification gains required decades of re-architecting factories around electric motors—an analogy for decentralized, right-sized AI use across many roles and processes. Practical steps discussed include: building an internal capability “ladder” for employee maturity, running adoption sprints and hackathons to engage the “crowd,” packaging learning into reusable artifacts (skills, repos, configs, sandboxes), and using guardrails/sandboxing to make autonomy safe. It also emphasizes unified, “tidy house” data environments to avoid fragmented access. Finally, the article connects diffusion to geopolitics: true “sovereign AI” depends on diffusion-friendly design, interoperability, and (properly understood) open-source architectures, not a single homogeneous global model. Overall, the central thesis is that AI adoption wins through patient embedding and incentives that reward sharing—rather than flashiest breakthroughs.