Master the Craft

Synopsis

Leadership is more like carpentry than calculus, and so is AI. You do not learn carpentry from a textbook; you learn it by cutting joints, ruining wood, adjusting, and cutting again. The dominant organisational response to AI — design a curriculum, deliver modules, certify completion — produces people who can talk about AI but cannot build with it, which is the definition of a failed pedagogy. Dreyfus and Dreyfus mapped why decades ago: formal instruction can carry a learner from novice to advanced beginner and no further. Everything past that requires situated practice on real problems under genuine uncertainty. This paper argues the practical consequence is architectural rather than educational. You do not produce AI craftspeople by improving the curriculum; you produce them by creating the conditions under which craft develops — daily practice, sandboxes where failure is safe, peer communities where knowledge circulates, and progressive challenge. The unit of learning is not the training course. It is the community of practice. It also names a bifurcation most organisations are not measuring: some people are atrophying as they offload thinking to AI, while others are accelerating as they co-create with it. The same tool produces both outcomes. The difference is not access — it is the discipline wrapped around the tool.

Key findings

• Formal instruction can take a learner from novice only as far as advanced beginner; competence and expertise arrive solely through situated practice on real problems under genuine uncertainty.
• Endoscopists exposed to AI-assisted colonoscopy for twelve weeks showed a 21% decline in adenoma detection when the AI was removed — deskilling measured in a clinical setting.
• Airline pilots whose manual flying was limited to minutes per flight showed degraded situational awareness, with up to 44% failing basic cognitive tasks in simulators.
• A World Bank randomised controlled trial in Edo State, Nigeria produced 0.31 standard deviations of learning gain — roughly two additional years of schooling — and closed the gender gap entirely. The critical variable was the structured pedagogy around the tool, not access to it.
• “Literacy” implies a minimum standard and “fluency” is just faster literacy spelled with an F; neither describes what organisations need, which is people who know their tools intimately and produce quality work through skill rather than luck.