AI coding tools are changing software engineering so quickly that they may be disrupting the traditional path through which junior developers become senior engineers. At QCon London, engineer Alasdair Allan argued that AI is eliminating many of the repetitive tasks through which inexperienced developers traditionally learned—writing basic code, debugging legacy systems and investigating production problems. At the same time, AI allows less-experienced developers to produce work beyond their current skill level, potentially masking gaps in understanding.
The concern is that AI still needs skilled supervision. Engineers must understand architecture, complexity, system behavior and failure modes to determine whether AI-generated code is actually correct. Allan argues that these skills traditionally developed through years of working with difficult codebases and real production incidents. If junior developers instead ask AI to explain legacy code or solve problems for them, they may reach answers faster but miss the learning process needed to develop engineering intuition.
The article also highlights evidence suggesting that AI's productivity benefits are more complicated than headline claims imply. Allan cites research in which experienced developers using AI were 19% slower despite believing they were faster, while another study found junior engineers using AI scored lower on mastery without completing tasks faster. He argues that AI works particularly well for short, well-defined tasks, but as complexity increases, the bottleneck shifts toward code review, architecture and judgment—the areas where experienced engineers provide the most value.
The biggest long-term concern is therefore the engineering talent pipeline. Companies may reduce entry-level hiring because AI can perform some junior-level work, but those junior engineers are precisely the people who would eventually become the senior engineers capable of supervising increasingly powerful AI systems. Allan's proposed response is to deliberately preserve structured learning: rotate junior engineers through fundamentals, measure understanding rather than just output speed, and treat AI as a tool rather than a teacher. The central warning is powerful: if companies optimize away the work through which engineers learn, they may eventually optimize away the supply of engineers capable of judging AI-generated work.
