The article argues that 'coding is solved' is a brain-dead narrative. The author, with experience building AI systems, contends that while LLMs can generate code, they fail at critical non-functional requirements like maintenance, reliability, and security. Even functional requirements remain unsolved.

The author identifies three product types that don't require code review: personal software, POCs, and weaponized AI—all high-risk tolerance scenarios. Production software in healthcare, finance, automotive, and defense demands accountability AI cannot provide. AI cannot be punished, held responsible, or suffer consequences, making it unsuitable for systems where mistakes cost lives or money.

The author criticizes proponents of 'coding is solved' for having little to show. They note LLMs succeed through feedback loops that hide errors, but struggle with logical reasoning and counting tasks. Coding remains one of the last areas for current LLM generation to take over due to its logical nature and compiler constraints.