I'll never forget the magical feeling I experienced when I wrote my first working program. I felt like a wizard, carefully casting spells to bend the computer to my will. For over 30 years, I've managed to keep that magical feeling alive and have done my best to stoke the same excitement in developers earlier in their careers. We get to be magicians!
A common topic of conversation amongst developers these days concerns the future of the profession. With AI coding agents continuing to advance, experienced developers are being given an incredible tool to accelerate their work. A big part of leveraging AI coding assistants is managing and directing their output to ensure that the code and architecture are of the right quality, and this fundamentally requires experience.
But what about early-career developers? What does AI mean for them? Respected experts have been noodling on these questions for a few years now, and consensus is forming around a critical conclusion — we must invest in hiring, training, and developing the next generation of programmers, not attempt to automate them away with AI.
Preceptorship
Last week, a new paper was published in the ACM library. Authored by Mark Russinovich — Azure's CTO and Technical Fellow — and Scott Hanselman, the paper is entitled Redefining the Software Engineering Profession for AI, and their thesis is simple: if we stop hiring junior developers because AI can write code, the profession's talent pipeline will collapse with disastrous consequences.
I've been managing engineering teams for over two decades, and this paper articulates something I've been feeling in my gut for the past year. The temptation is real. When AI coding tools can generate working code in seconds, the ROI math on hiring an early-in-career developer who needs months of mentoring to reach productivity seems obvious. Why invest in humans when you can invest in tokens? In my view, that math is dangerously shortsighted, and Russinovich and Hanselman explain exactly why.
The senior engineers who make AI tools effective didn't spring into existence fully formed. They were once the junior developers someone invested in. They built their intuition through years of debugging production outages, making architectural decisions that aged poorly, and learning from mentors who took the time to explain why, not just what. That hard-won judgment is precisely what makes them capable of reviewing AI-generated code, catching subtle bugs, and prompting effectively.
The paper proposes what they call "preceptorship at scale" — borrowing from medicine, where new doctors train under experienced physicians in a structured, deliberate system. The analogy is apt. You wouldn't let an AI diagnose patients without training new doctors, no matter how good the AI got. The human expertise is what makes the AI useful.
Intention is Required for Invention
I've been thinking about this in the context of my own team. We use AI coding assistants daily — Claude, Kiro, Cursor, etc. They're incredible force multipliers. But the engineers who get the most out of these tools are the ones with deep foundational knowledge. They know why the code works, not just that it works. They understand system architecture in ways that let them evaluate whether the AI's suggestion is brilliant or subtly catastrophic.
But I have good news — AI could actually make mentorship more effective, not less. Imagine an early-career developer working alongside both a human mentor and an AI assistant. The AI handles the boilerplate, the syntax questions, the "how do I write a unit test for this" moments — freeing the mentor to focus on architecture, design patterns, judgment calls. The stuff that takes years to develop. The stuff AI still struggles with.
But this only works if organizations make it intentional. The default path — stop hiring juniors, lean on AI, milk your existing senior talent until they burn out or retire — leads to a future where there's nobody left who understands the systems deeply enough to guide the AI effectively. It's a talent debt that compounds faster than technical debt.
My takeaway for tech leaders: resist the urge to cut your early-career pipeline. Rethink how you develop talent. Use AI to accelerate learning, not to eliminate learners. The companies that figure out this balance will have a massive competitive advantage in five years. The ones that don't will be scrambling to hire experienced engineers from a pool that stopped growing.
Investing in people has always been the right long-term bet. AI doesn't change that and I doubt it ever will.
