The difference is that you'll get a different "ZX Spectrum GUI" each time with an LLM, but the same result with a compiler. The actual work in both cases is no

The difference is that you'll get a different "ZX Spectrum GUI" each time with an LLM, but the same result with a compiler. The actual work in both cases is not to write the source code, but to figure out and describe exactly what you want. In one case (traditional programming) this specificion happens directly in non-ambigious source code, in the other case (steering an LLM) you specify via an imprecise human language. Quite often the human language specification is actually "good enough" to get what you wanted, but the more you need to go into specific details, the more the difference to traditional coding disappears.