Interesting take. I hadn’t peeled back the skin on this domain of work, yet.

My educational background is in creative writing (poetry), though I’ve found that programming is at least as good (probably better) at scratching the creative itch that led me to poetry in the first place. I’m obliged to wonder why, and I suspect part of this has to do with how relatively tricky it is to deeply communicate with and satisfy even one human vs something like a compiler/interpreter and test suite.

You can try to solve for this by trading feedback in workshops or with friends, or being successful enough to have stable editor/agent relationships, but it’s still the case that the impression of any one person is a poor proxy for your (ideal) audience (and that person’s feedback might greatly differ from day to day or year to year). It’s also the case that a writer can’t author and share millions of poems a day to discover what works without quickly burning through everyone who’ll give them the time of day.

I guess where I’m a little unsure around this argument is that prompting strikes me a (~new) kind of communication with an (often but not necessarily) probabilistic oracle that can (depending on your resources) tirelessly give near-instant feedback. Prompt engineering smells like something humans will be good at early on while there’s a lot of low-hanging fruit and no/few automated tools, but that in the long run we’ll be using some kind of automated combinatorial/fuzzing approach to pick at LLMs and systems built on them even if just to systematize learning about and documenting their capabilities.