Podcast: How Will We Train Developers If AI Does the Routine Work: A Conversation with Scott Hanselman

Details

In this InfoQ podcast, Scott Hanselman (Microsoft) and Michael Stiefel examine a structural hole opening in the engineering talent pipeline: AI now handles the low-stakes, high-repetition work (tests, config files, boilerplate, simple features) that has always been the training ground for junior developers.\n\nHanselman's proposed fix borrows from nursing: a preceptorship model where designated senior engineers serve as trainers full-time, evaluated on how many developers they grow rather than how much code they produce. He argues this requires a deliberate economic decision by companies to invest in human capital even when AI can temporarily cover the output gap.\n\nOn the capability split, Hanselman draws a line: AI agents are strong at generating features in isolation, but weak at software architecture because they lack the wider organizational and domain context that shapes good design. Engineers become reviewers, coordinators, and context-setters rather than primary authors.\n\nThe conversation extends to isolation risks: remote work, social media, and AI-assisted coding compound to reduce the informal learning and human connection that junior engineers have historically absorbed by working alongside seniors. Hanselman calls for intentional in-person interaction as a countermeasure. The podcast does not name specific tools, timelines, or data points, keeping the discussion at a conceptual level.