Unpopular opinion: prompt engineering is a dead end job and we all know it
I spent 6 weeks last spring obsessing over system prompts for a support bot at my company in Austin, and the model still broke every time a user asked two questions in one message. Then we swapped in a newer model with a simple one line instruction and it beat my best prompt by 30 percent on our test set. The whole field of writing clever little word tricks is just a patch for gaps that model updates close every few months. So when I see people selling $500 prompt courses, I cringe, because the skill has a shelf life of maybe a year. What are you actually building that survives the next model drop?
Six weeks of tweaking prompts and a one liner beat it by 30 percent? That would sting so bad I'd just close my laptop and walk away from the whole thing. The part about selling $500 courses is what gets me, since you're basically paying to learn tricks that expire before you finish the lessons. Whatever survives the next model drop is probably just knowing what you want the thing to do, not how to word it.