Quick Buzz Feed

How Artificial Intelligence Disrupts Engineering Progression

Gary Lloyd | Aug 13,26 | 01:34 EST

Technology

AI is fundamentally altering career paths in engineering by removing critical learning opportunities at various experience levels, while simultaneously empowering individuals to perform tasks beyond their current expertise. Alasdair Allan highlighted this disruption in his talk, "Engineering Progression When AI Ate the Middle," at QCon London. The industry is seeing a decline in the number of junior developers entering the workforce, and AI's influence is contributing to a slowdown in entry-level hiring. Allan noted that AI is now responsible for generating a substantial amount of code, which has dramatically reshaped the role of a developer. However, he emphasized that writing code was never the sole purpose of the engineering profession. Effective utilization of AI still necessitates human supervision, which in turn requires strong underlying coding skills. Allan poses a critical question: if AI takes over the work that traditionally served to train new engineers, where will the next generation of skilled engineers emerge from? He suggests that the immediate productivity gains from AI might come at the expense of developing the essential skills needed to validate AI-generated code, potentially stunting the skill development of junior engineers. The intuitive pattern recognition, understanding of system architecture, identification of hidden complexities, and foresight into scalability issues—qualities developed by veteran coders—are no longer being acquired through traditional means. Junior engineers, Allan observes, are not building the necessary intuition to discern the correctness and suitability of the code they encounter; instead of deep diving into legacy codebases or debugging complex production incidents, they are increasingly relying on AI agents for summaries and solutions. A significant portion of engineering involves maintaining 'blackfield' systems—legacy platforms under heavy load, slated for deprecation but never getting the attention needed. These systems often lack proper documentation, or their specifications have been rendered obsolete by years of unrecorded decisions, with business logic embedded in conditions that predate anyone who understands them. While AI agents can read code and documentation, they cannot interpret production behavior—they lack the ability to analyze years of request patterns to understand which code paths are critical in ways the code itself doesn't explicitly state. Allan pointed out a slowdown in the hiring of young workers in AI-exposed occupations, with no corresponding decrease in work for experienced professionals over 25. This trend translates to fewer junior developers joining the industry, stating that "People weren’t being sacked; they just weren’t being hired in the first place." He concludes that "AI stunts skill formation, AI transforms work into supervision, and now AI slows hiring at the entry level." Organizations rushing to integrate AI coding tools risk undermining the very pipeline that produces individuals capable of effectively supervising those tools. The ability to construct accurate contexts for AI agents stems from years of hands-on experience and accumulated craft knowledge, a pipeline that Allan argues is currently fracturing. He likens the situation to medical residencies, where "scut work" is essential for teaching and developing judgment, not just for efficiency. Allan believes that companies prioritizing this foundational, often less efficient, learning will be the ones that retain senior developers a decade from now, as others might face a shortage due to early retirements among their experienced staff.

Conclusions from studies on AI's impact on development time and code quality

Research findings are contradictory regarding AI's impact on development time and code quality. Some studies indicate experienced developers are slower with AI, despite believing they are faster. Junior engineers using AI have shown lower mastery scores without completing tasks more quickly, suggesting they sacrifice learning for no tangible benefit. While AI proves useful for brief, well-defined tasks, its advantages diminish with increasing complexity, shifting the bottleneck to crucial code review stages that demand senior judgment. Many AI-generated pull requests in real open-source projects were unmergeable, even when passing automated tests, highlighting AI's proficiency in functionality over craftsmanship.

How computer programmers' work changes with AI tools

Programming is transitioning into a supervisory role. AI handles much of the code generation, making humans responsible for judgment and validation. Only a small fraction of daily engineering work can be fully delegated to AI. A major concern is the growing trend of engineers seeking answers from AI instead of consulting colleagues, thereby circumventing the invaluable incidental learning and skill development traditionally gained through collaborative problem-solving and mentorship. The existing career pipeline, designed around coding, has yet to adapt to this new supervisory paradigm for systems.

Suggestions for building engineers in the AI era

While a definitive solution for building engineers in the AI era is still evolving, immediate steps can be taken. These include establishing structured learning paths with intentional rotations through fundamental tasks, similar to medical residencies, to cultivate essential judgment. The focus should shift from measuring task velocity to assessing genuine understanding. It is crucial to observe engineers' thought processes rather than solely their output, and to treat organizational context as vital infrastructure. Documentation should be crafted to onboard senior engineers, assuming coding proficiency but providing necessary codebase context. Ultimately, setting realistic expectations is key: AI is an assistive tool, not a teacher, and using it to bypass foundational understanding risks long-term skill deficits.