tl;dr: A veteran engineer pushes back against the "coding is solved" narrative, arguing that LLMs handle code generation but fail at the harder parts of software engineering: maintenance, reliability, security, and accountability—all critical for regulated industries like healthcare, finance, and aviation. The author contends that AI can't be held accountable, produces stochastic output that's hard to verify at scale, and that executives measuring token usage as productivity are setting up costly failures. Coding is deemed suitable for POCs, personal software, and throwaway automation, but professional software still requires human understanding and ownership.