You have 1 article left to read this month before you need to register a free LeadDev.com account.
Estimated reading time: 5 minutes
Key takeaways:
- Two Microsoft engineers have proposed that the industry adopts a nursing-style preceptorship model, where juniors and seniors work as equals to deconstruct and debug AI-generated code over a year-long period.
- As AI automates grunt work, the engineer’s role is shifting from manual coding to being a “connoisseur of output”.
- Transitioning to this model requires companies to view mentorship as a method for retaining talent, even if it initially slows down short-term development speed.
Software preceptorships could redefine the relationship among junior developers, senior engineers, and AI.
A piece of research led by engineers at Microsoft has proposed a new way to train the next generation of software engineers, as AI threatens their livelihoods. It borrows a practice from the nursing profession: the preceptorship.
“The path forward is a culture of preceptorship at scale,” write Microsoft Azure CTO Mark Russinovich and VP of developer community, Scott Hanselman, in the paper, titled: Redefining the Software Engineering Profession for AI.
“We must enable senior mentorship with AI systems that capture reasoning, surface misconceptions, and turn daily work into teachable moments” for junior developers.
The widening seniority gap
The next generation of AI coding tools could very well eliminate all the gruntwork junior devs perform today, though chances are we will still need someone to review that code, currently the role of the wizened senior engineer.
Only skilled engineers will be able to catch the subtle bugs, architectural flaws, and vulnerabilities in AI-generated code. “Human oversight, critical thinking, and domain knowledge are indispensable for both correcting errors and driving innovation as technology progresses,” the pair write.
But where will tomorrow’s senior engineers come from if companies stop hiring junior developers? Preceptorship pairs junior software developers with senior mentors. Together, they build with AI, then debug the result. Using this model, learning the stack remains essential, but real knowledge will come from “making the rounds” or figuring out the edge cases where some AI code doesn’t work as expected.
The process would require companies, much like hospitals today, to make “growth an explicit organizational goal,” the Microsoft authors write.
Your inbox, upgraded.
Receive weekly engineering insights to level up your leadership approach.
“Connoisseurs of output”
In a sense, agentic coding assistants provide an embarrassment of riches. In the words of the authors, agents can “interpret goals, reason across repositories, and iteratively generate, test, and refine code” in a fraction of time it would take human coders to do the same. But, their work still needs review and redirection.
AI is just another jump up in abstraction, much like the compiler was back in its day, Grady Booch, the now-retired UML creator and former IBM chief scientist said as much in a recent InfoQ podcast. The compiler automates a lot of low-level work, but a dev still needs to channel the results into a shippable product. Likewise, AI allows humans to think at a higher level of abstraction. It’s great at automation, but it still needs guidance.
The Microsoft paper quotes Ethan Mollick, Associate Professor of Management at the Wharton School of the University of Pennsylvania, who writes: “We need to become connoisseurs of output rather than process. We need to curate and select among the outputs the AI provides.”
Some developers at Spotify, for instance, have largely stopped writing code by hand, and instead are using generative AI to produce new features and debug issues. The services’ “best” developers “have not written a single line of code since December,” the co-CEO Gustav Söderström stated during a February earnings call. Instead, they inspect the AI work, making any necessary changes, according to TechCrunch.
As Burke Holland, a Microsoft GitHub Copilot developer, said on the Changelog podcast, “you still need to be a high-performing developer to be able to ship.”
Even with AI coding tools as powerful as Claude Opus, “you need to understand a lot about architecture and security and how an app gets to production. You still need to know a substantial amount about what you’re doing.”
Anthropic itself has dozens of job openings for senior developers. “Someone has to prompt the Claudes, talk to customers, coordinate with other teams, decide what to build next,” admits Anthropic’s Boris Cheney in a comment on X.
More like this
Not just an internship
Hanselman and Russinovich got the idea of a preceptorship from Hanselman’s wife, who, as a student nurse, went through a preceptor program at a hospital. Originally, Russinovich, discussing the issue with CEO Satya Nadella, envisioned the idea of an internship of some sort.
That didn’t quite get it for Hanselman, who saw traditional internships as placing interns on unequal footing with the mentors. Instead, the preceptor relationship should be of equals, allowing the preceptee to freely ask the preceptor questions.
“Becoming a nurse doesn’t mean you’ve just “done training.’” Hanselman says in an interview with LeadDev. Just as a nurse needs to prove clinical readiness, engineers need to do the same to earn the title.
The authors envision the software engineering preceptorship would typically last a year or even longer. The preceptor and preceptee would “experiment freely,” through a series of small iterative projects. Instead of authoring code, they would direct agents to produce the code, then deconstruct the results.
Getting software-defined organizations to adopt the preceptorships will be a challenge, the pair admit. They would undoubtedly slow developer velocity, as more of their time would be dedicated to mentorships. It could be a hard sell to investors. Software organizations would have to be convinced to borrow a page from hospitals, who value preceptorships as a cost-avoidance measure to retain expensive-to-replace talent.
Hanselman and Russinovich are considering writing a follow up paper that focuses on implementation.
Preceptorships beyond AI
Honeycomb CTO and co-founder Charity Majors has long warned the industry about the pending crisis of not fielding enough junior developers, even before the onset of AI.
New York • September 15-16, 2026
Speakers Gergely Orosz, Will Larson and Frances Thai confirmed 🙌
Responding to the Microsoft paper on X, she notes that, “At every place that I have seen start hiring junior engineers in the last few years, that charge was led and lobbied for by senior engineers.”
It’s the senior engineers that saw the need to keep the institutional knowledge intact. This tracks with what Hanselman says as well.
“In the waning years of my career the only thing that matters is that I pass knowledge on to the next generation. I made it 30+ years into this career without quitting, giving up, or being booted out, and [junior developers] deserve the same, regardless of AI.”
