Until the moment we start bragging about how many lines of code LLMs are saving us, we're walking in the wrong direction. Your programs, designs and architectures is supposed to get better, not add even more boilerplate just because you can produce it faster...
"You go to IPO with the AI you have, not the AI you might wish you have." -- Donald Rumsfeld
So, right now it's a verbose code generator.
But post-IPO it will be wonderful - sentient, self-improving (recursively, iteratively, asymptotically), full of loving grace.