Training Tomorrow's Engineers When AI Handles the Work: Scott Hanselman on a New Model

As artificial intelligence tools grow capable of generating software features, the tech industry faces a fundamental question: how do you train the next generation of engineers when the routine work that built their skills no longer exists?

In a recent conversation on the Architects Podcast, Scott Hanselman, a vice president at Microsoft GitHub, explored this challenge alongside host Michael Stiefel. Hanselman's answer draws from an unexpected source—the nursing profession—and calls for a wholesale rethinking of how software companies develop talent.

The Preceptorship Model: Lessons From Nursing#

Hanselman and Mark Russinovich, in a piece published in Communications of the ACM, argued that the software industry should adopt a preceptorship framework. Rather than having junior developers learn through doing tedious work alongside busy senior engineers, teams would employ dedicated preceptors—professionals whose sole responsibility is training new engineers.

The distinction is crucial. In nursing, preceptors are formally recognized roles evaluated not on how much work they personally ship, but on whether they successfully develop competent nurses. By contrast, software teams typically assign mentoring duties to architects or tech leads already stretched across projects. Those leaders face little accountability if junior staff fail to advance; they simply hire more interns and repeat the cycle.

"Their job is not to ship code, but to ship engineers," Hanselman explained, describing how preceptor-led teams would pair experienced engineers alongside learners for pair programming and collaborative problem-solving—ensuring both product output and human growth. A company's annual review of a preceptor would center on whether they've developed capable engineers, not on their personal code commits.

AI's Blindspot: The Big Picture#

While AI excels at generating individual features, Hanselman highlighted a critical gap: AI struggles with software architecture. He cited his own experience watching an AI tool gradually construct a "god object"—a bloated, overgeneralized data structure that would become an architectural liability. AI agents tend to be myopic, missing the larger systems context that experienced architects naturally consider.

This limitation underscores why human oversight remains essential. Engineers must act as reviewers and coordinators, maintaining judgment over AI-generated code and preventing short-term solutions from calcifying into long-term technical debt.