Doesn't 'prompt engineering' sound a bit ridiculous now?

Just two years ago, a lot of people were saying that "those who write good prompts get results," but these days it really seems like the models just take over and run with it themselves.

Of course, the ability to organize context and feed it in is still important. But isn't that closer to simply "understanding the work well" than to a prompting technique? After all, you don't need to add instructions like "think step by step" anymore—they do just fine without them.

In the end, what remains is 1) the ability to clearly define the problem and 2) the domain knowledge to verify the output. I think it comes down to those two. The know-how for using tools will keep getting commoditized.