What exactly is the hard part they’re describing? Taste? Picking a stack? Noticing drift?

Seems like the “harder and vaguer thing” would be explaining to someone why that’s even necessary, or why your particular taste or stack of choice are any better than what the AI would choose itself.

Kinda silly when the whole point of the piece is “don’t invent fake things to justify your work”.

Whenever possible, I ask leading questions rather than telling it explicitly what to do. I give it a problem and ask it what changes would be needed and if there are any complications. If I don't like the plan it comes up with, then I point out things and ask other questions. Make the agent do its own "thinking."

very often I find people who complain AI is not good yet for their coding projects

mostly because they try it looking for a flaw to discard it in a confirmation bias/change resistance kind of way as soon as possible

had to put our team to use it obligatorily for some weeks, after that each one shared a lot of skills, workflows, plugins and really got into it as an accelerator to complete their ideas