Am I the only one appalled by the audacity to call the writing of vague instructions for an AI "engineering"? However many levels of indirection there might be at play.
1. Building engines from scratch, i.e. something that converts a thing from X to Y consistently.
2. Knowing enough about an engine to get it to do what you want it to do, as well as maintaining it.
Prompt engineering initially started from the latter, back before ChatGPT and stuff which you'd just give instructions to.
Engineers have to be familiar enough to know how it works, what it's strong at, what it isn't, things like margin of failure, all that stuff. A lot of it is just prompt alchemy, I guess, but the lower part of this article has some juicy reverse engineering.
"oh, it doesn't see this condition, so if I help it a little bit, then... Ah now it unrolled, and inlined, but accidentally trashed instruction cache, how will I convince it not to..."
Or maybe it's just me associating engineering with hard science.