Programming languages for AI

Mark Seemann argues that if LLMs are writing all code, the choice of programming language still matters — not for human readability, but for portability, security, and verifiability. He critiques the notion that LLMs could just generate machine code or C, pointing out the safety and correctness risks. Instead, he advocates for languages with strong static type systems, algebraic data types, and ideally dependent or refinement types. He suggests an AI-first language should be designed for maximum verifiability rather than ease of writing, coining the idea of 'design by inconvenience' for machine-generated code.