I don't think this comparison holds up. With a higher-level language, the material you're building with is a formal description of the software, which can be fed back into a compiler to get a deterministic outcome.
With an LLM, you put in a high-level description, and then check in the "machine code" (generated code).