Learning Programming in an Age of LLMs

A recent letter prompted me to ponder the implications of learning programming amid the rise of large language models (LLMs). While I acknowledge the uncertain nature of this technological transition, I remain fascinated by the potential benefits and drawbacks. Having built a complex TypeScript/JavaScript system using AI tools, I now find myself struggling to maintain control over my own creation.

When everything functions smoothly, the gap between my understanding and the system's capabilities is easy to overlook. However, when issues arise, that gap becomes painfully evident. At times, I find myself questioning whether I've merely constructed an impressive façade rather than a fully comprehended product. While I am not opposed to AI, I am uncertain about the optimal relationship between humans and these sophisticated systems.

I find myself leaning towards disliking AI, even as I recognize its potential and occasional usefulness. My skepticism is rooted in my socio-economic position and concerns about mass unemployment among knowledge workers, which could have severe economic repercussions. As an economist, I am aware that new jobs may emerge, but the transition is not always smooth, as seen in historical examples like the introduction of the stocking frame or computers.

Nevertheless, I remain hopeful that there will be opportunities for future generations, even if they differ from those we know today. After reflecting on the matter, I realize that learning programming fundamentals before LLMs was invaluable. Those skills served me well for three decades, and while I aspire to continue programming for another ten years, I also understand the importance of preparing the next generation for an uncertain and rapidly evolving landscape.