The conversation about how we train the next generation of software engineers keeps circling back to the same uncomfortable question: if AI agents are now doing the routine work that junior developers once learned from, what exactly are they learning? In a recent podcast, Scott Hanselman proposes that the industry borrow a page from nursing education and adopt a preceptorship model, where experienced engineers mentor novices through structured, hands-on practice rather than expecting them to absorb skills through osmosis. It is a sharp observation, and one that deserves more than a nod. The old assumptions about career progression are breaking down, and pretending otherwise leaves us with a generation of developers who can prompt an AI but cannot reason about the code it produces.
This is not just a training problem. It is a signal that the entire skill stack for software development is shifting underneath us. We have written before about how Navigating AI/ML Job Requirements: A Shift in Expected Skills reflects a market that no longer knows what it wants from engineers, and Hanselman's point cuts to the same nerve. If the routine work is automated, the entry-level role stops being a place to make mistakes and learn the basics. Instead, it becomes a place where you are expected to understand the trade-offs of solutions you did not design. That is a heavy ask for someone who has never debugged a memory leak or reasoned through a race condition. A preceptorship model, with its emphasis on deliberate supervision and progressive responsibility, is a practical answer to a real gap. It is not glamorous, but it is honest.
What makes this idea compelling is that it reframes the goal of early-career development. Instead of measuring progress by tickets closed or features shipped, the focus shifts to judgment: knowing when to trust the AI, when to question it, and when to step in and write the code yourself. That is a harder skill to teach, and it will not happen by accident. We have also touched on how Verify Your AI's Understanding: A Simple Check for Tax Season highlights the importance of testing whether a model truly understands what it is doing, and the same principle applies here. A junior developer who cannot verify what the AI produces is not a developer; they are an operator. The preceptorship model forces that verification skill into the center of the learning process, which is exactly where it belongs.
The open question is whether the industry has the patience for this. Preceptorships require senior engineers to spend real time teaching, and that time is already scarce. But the alternative is a workforce that is technically fluent and conceptually hollow. We would tell any reader who is worried about their own path: do not wait for a formal program. Seek out environments where someone reviews your work and explains the reasoning behind the review. Ask for the kind of feedback that hurts a little. And if you are the senior in the room, take the junior under your wing not because it is efficient, but because it is the only way the craft survives. The routine work is gone. The teaching should not be.