Generative AI is transforming software development at remarkable speed. But in a new paper written by two Microsoft executives, Azure CTO Mark Russinovich and Scott Hanselman, VP of the developer community, argue that the AI technology also risks eroding the profession’s training pipeline.
In their paper, Redefining the Software Engineering Profession for AI, published in the journal ACM, the authors contend that agentic coding assistants disproportionately amplify experienced engineers while creating new challenges for junior developers. The result, they write, is an economic incentive to hire seasoned staff and use AI for lower level work, a shift that could leave the industry without its next generation of technical leaders.
The paper argues that organizations must continue hiring junior developers even if they initially reduce short-term output. More importantly, companies should design systems that make developer growth a core objective, not a byproduct of production.
As developers know well, agentic coding assistants now go well beyond autocomplete. These systems interpret high-level goals, reason across repositories and generate and refine code. Engineers often shift from writing code to directing the system, reviewing its output and integrating it into a coherent whole.
Yet the paper stresses that programming and software engineering are not synonymous. AI tools can produce large volumes of functional code, but they often fail in important ways. The authors describe examples in which agents introduced inefficient algorithms, duplicated logic across codebases, ignored crashes as irrelevant to a task or applied quick fixes that masked deeper flaws.
In one case, an AI assistant attempted to resolve a race condition by inserting a delay, effectively suppressing visible symptoms without addressing the underlying synchronization bug. A developer lacking experience in concurrency might view this as a valid solution. Only an engineer with experience and familiarity with this type of problem could confidently reject it and guide the system toward a durable fix.
AI and Hiring
This dynamic, the authors argue, is already influencing hiring patterns. Citing labor data and academic research, they note evidence that roles for younger workers in AI-exposed occupations have declined relative to senior positions since the release of advanced gen AI tools. A Harvard study described the trend as “seniority-biased technological change,” in which AI increases the value of existing expertise while lowering opportunities for newcomers to build it.
The long-term concern is what Russinovich and Hanselman call a narrowing pyramid. Traditionally, teams of junior engineers handled simpler tasks, gradually developing judgment and workflow instincts. Some advanced into managerial roles, sustaining the profession’s depth. If companies reduce early-career hiring, that progression stalls.
A Proposed Solution
The authors’ proposed remedy is a formalized preceptor model. Senior engineers would take responsibility for mentoring groups of early-career developers within real product teams. AI tools would be configured to support learning, not just throughput. For example, by defaulting to Socratic questioning before generating code, surfacing reasoning steps and allowing mentors to review interaction logs.
The goal is to convert what the authors describe as AI drag into deliberate capacity building. Junior engineers would participate in debugging, design trade-offs and prompt refinement alongside mentors, observing how expertise shapes interaction with AI systems.
The issues around AI in software development spark plenty of heated debate. Some experts say that AI tools are improving rapidly and may actually narrow the experience gap. Others point out that Microsoft itself has reduced engineering headcount in recent years. The authors, for their part, frame their argument as a professional responsibility rather than a corporate directive (and in any case, the paper is the authors’ viewpoint, not a Microsoft corporate release).
Ultimately, the two Microsoft executives argue that the defining question for software engineering is not how much code machines can produce, but how effectively humans learn to reason with them. Without sustained investment in early-career talent, today’s AI based gains could give way to tomorrow’s skills shortage.
