Coding Is Not Solved – Alex Ewerlöf Notes (blog.alexewerlof.com)

Alex Ewerlöf argues that the perception that "coding is solved" through AI and large language models (LLMs) is a misleading narrative. He emphasizes that while LLMs can generate code, they struggle with non-functional requirements (NFRs) such as maintenance, reliability, and security—importantly highlighting that most software development, especially in critical sectors like healthcare and finance, demands high accountability that AI cannot provide. Ewerlöf cites his experience as a seasoned developer to caution against over-relying on AI, suggesting that while LLMs may accelerate development in low-risk scenarios like personal projects or proofs of concept, they are fundamentally limited when applied in more serious contexts. Ewerlöf underscores the inherent logical challenges faced by LLMs, which are probabilistic and often produce flawed outputs unless meticulously supervised and corrected. He warns that belief in LLMs' capabilities reflects a lack of understanding of coding and its complexities. While recognizing that AI technologies have transformative potential in software development, he calls for a balanced approach—advocating for the maintenance of high coding standards and critical thinking among developers to avoid rendering their skills obsolete as AI continues to evolve.