I'm totally in agreement as far as the current state of NLP and AI goes, but all bets may be off as these fields mature. For example, I have a client that gives me a very detailed specification for a simple iPhone app in a few English paragraphs, along with storyboards, wireframes, and similar apps. I bang out an alpha in a week or two that's mostly faithful to her conception. She writes up some minor changes she'd like, I implement them, get back to her, and on we go until the app is close enough to what she had in mind.
In this case, I'm essentially a really slow compiler turning English language into machine language. Who can say to what extent machines will be able to perform this function in the future?
I'm obviously oversimplifying, and an AI-driven English->asm compiler comparable to my godlike programming abilities is at the far, far end of the technology gradient from now to Lieutenant Data, but it doesn't seem far fetched to have such compilers be "good enough" for designers to create certain MVPs and such without programmer intervention, in the next decade or so.
In this case, I'm essentially a really slow compiler turning English language into machine language. Who can say to what extent machines will be able to perform this function in the future?