I think you are critically (I'm not saying intentionally) missing the middle ground in your assessment. Large generative models are great for repetitive, homogenous tasks like surveillance, speech recognition, assembly line quality control, etc.
If I try to look at it from your point of view, regarding creative tasks like developing new software, you have one side claiming that you can make hammers out of chocolate and throw away all your proper ones, and another that's portrayed as being unreasonable for opposing that.