Situation: every streaming service recommends the same "popular now" to everyone, then buries the thing you'd actually love. It doesn't know your taste, it doesn't know what you've already seen, and it has no idea what you'd never watch.
The method applied: model the self. A structured record of what I've watched, what I loved, and — crucially — my anti-interests (the genres and tropes I'd never pick) becomes the input. The AI recommends from that modeled preference instead of a crowd average. The "not" list does as much work as the "yes" list: telling the system what to rule out sharpens every recommendation that remains.
Why it's a use case: it's the smallest, clearest demonstration of two named concepts working together — a modeled self (the AI needs a structured picture of you to be a real assistant) and anti-interests (defining what you're not is an act of focus). Same pattern as a business filtering leads, content, or suppliers: model the preference, encode the vetoes, let the structure do the sorting.
What changes when structure is added: recommendations go from generic to mine — because the model has a real, structured preference to reason over instead of a guess.