What Should We Learn? What LLMs Taught Me: Understanding Isn't Producing (dennysora.me)

In a reflective piece, the author discusses the implications of using large language models (LLMs) like Opus 5.5 for coding and learning. They argue that while AI-generated code may often be superior to the work of many human developers, overreliance on these tools risks diminishing foundational coding skills and the understanding of architectural principles. As developers increasingly shift from traditional coding roles to positions focused on feature development, there’s a concern that many may lack critical design abilities and deeper insights into their code, leading to a disconnect between product functionality and sound technical practices. The author also shares personal experiences that highlight the cognitive effects of leaning heavily on LLMs for tasks like writing and programming. They draw parallels between learning to play an instrument and using AI tools, suggesting that merely consuming well-structured output does not equate to genuine understanding or ability. This raises a significant question for the AI and ML community: as LLMs evolve and integration expands, what core skills and learning approaches should individuals prioritize to ensure they are not left behind in a rapidly changing landscape? The piece ultimately encourages a reevaluation of learning strategies to reconcile the strengths of AI with the need for human insight and structured thinking.