Lately I've been using a few different AI Lens-style features — the kind that let you point a camera at something and get a response back. The first time, it's fun. By the third time, it feels thin.
The natural question is "what went wrong with the feature?" But I think that's the wrong question. The real question is simpler and older than AI: why would anyone keep coming back to this at all?
Paying for something is just the strongest version of that question. Even a free product has to answer it. I've been watching a free AI camera app where the interesting question is not monetization yet. It is whether first-use curiosity can become a reason to return. That's a retention problem, not a pricing problem, and the same three things explain it whether money changes hands or not.
Utility
Does it clearly help? Faster, findable, decidable, buildable, fewer mistakes. Translate this menu, identify this plant, read this label — these are real utility. But utility alone rarely earns loyalty, because utility is copyable. Whatever a small AI feature can do, a general assistant already open in another tab can usually do too. Utility gets you tried. It doesn't get you kept.
Accumulation
Does it become more yours the more you use it? History, preferences, photos, belongings, context, judgment — the kind of thing that quietly makes leaving expensive. A wardrobe app that remembers what you own. A reading log that knows what you've finished. A tool that has, over time, absorbed something about you that would be annoying to rebuild somewhere else.
This is closer to what separates a feature you open once from one you open out of habit. But accumulation by itself isn't enough either — a pile of stored photos with nothing done with them is just clutter with a login.
Expertise
Does it show you something you couldn't see yourself? Put language to something you couldn't quite articulate? This is where taste, interpretation, and judgment live. Point an AI Lens at a genuinely interesting object and it can say something worth reading — but that's borrowed expertise, borrowing the user's own eye for what's interesting. The category only holds up when the interpretation itself is doing real work: naming a style, identifying a pattern in the object, explaining why something feels coherent or off, or giving the user a judgment they could not reach from surface description alone.
The risk is that most interpretive features are quietly designed for an imagined user with real taste — someone who'd point a camera at art, architecture, a considered outfit. The actual majority of users point it at leftovers and their cat. The first few times, the AI's warmth feels like being understood. But once it keeps producing the same eloquent commentary on inputs that have nothing to say, the hollowness becomes visible — and that's worse than boring. It's the moment "you get me" turns into "you don't get anything," and that's exactly when people stop opening the app.
Where this leaves most AI features
Most single-purpose AI Lens features are pure utility with no accumulation and no real expertise behind them — which is exactly why they get tried once and abandoned. The features that hold up over time tend to stack at least two of these together: a tool that accumulates your history and does something useful with it, or one that adds real judgment on top of what it remembers about you.
None of this requires payment to be true. A free product that people keep returning to has usually already solved this. A paid product that hasn't solved this is just charging for a first impression.