Most founders scope too big. The first version of an AI product should do one thing a human currently does by hand, and do it well enough that someone would pay for it. Everything else is a distraction until that one thing works.
Start by writing the single sentence a customer would say after using it. If you cannot write that sentence, you are not ready to build. If you can, that sentence is your spec.
From there, the build is short. A focused engineering crew can take a clear, single-outcome scope from idea to a working product in a few weeks, because there is nothing to argue about. The scope is the sentence.
Resist the urge to add a second feature before the first one has a user. The fastest path to a real product is one honest outcome, shipped, then watched.