I think this is falling for the same illusion as generative art, since programming has a huge creative component. Useful code tends to solve unique problems, otherwise you can copy/paste or import a library, just as interesting art is relatively unique and has its own identity. Generated art tends to have an average feel to it, despite its immaculate rendering. When you push stable diffusion to produce creative works, it struggles, just like when you prompt for code it will output generic solutions, if not broken code.
Agentic flows might have better outputs but there's a creativity component that seems intracktable at the moment, which is necessary to produce actual useful code. Absent this creativity component, it might serve as a good prototype tool for non-technical people, or a better autocomplete as another IDE tool.