In my experience, usually the problem with a non-technical person isn't that they can't implement their ideas. It's that even if they could, their ideas are unworkable for a hundred reasons that they can't begin to understand. So why would the outcome be any different with AI? Even if the AI is a super-intelligent system designer and tells them all the flaws with their thinking, they'll tell the AI to shut up and get to work just as they'd do with a software engineer.

I actually think that for the foreseeable future, the opposite is more likely. A technical person will be required to get good technical results out of AI, whereas a lot of the key growth-related grunt-work that a good non-technical founder takes on will become easier and easier for a skilled engineer to automate.