Restraint Is the Product

I’ve been using AI products in commerce for a year, and the same thing happens every time. I type something. It returns 200 results. I scroll. I lose the thread. I close the tab.¶

If you hired a stylist for a wedding and they sent you 500 suits, you’d ask why you hired them. The job isn’t to surface every option. It’s to ask the right question, narrow, bring you five.¶

Most AI products I’m using right now do the opposite. The optimization is throughput. Better ranker. Faster results. Broader catalog. The metric is how many things we surface. The experience is overwhelm.¶

We’re building search engines with manners.¶

The thing a stylist does that an algorithm doesn’t is ask a clarifying question before working. Black tie or cocktail? Are you in the wedding party or just attending? Beach ceremony or ballroom? Those questions aren’t friction. They’re the actual service. Without them, the stylist is a Pinterest board on legs.¶

Most AI consumer products treat the clarifying question as a UX cost. Fewer steps to result. Simpler input. As if the user wanted the result, when what the user actually wanted was someone who understood enough to bring fewer, better options.¶

The clarifying question is the feature. Restraint is the product.¶

The companies winning consumer engagement in 2026 aren’t the ones with the biggest catalog or the fastest ranker. They’re the ones that figured out how to bring you five things instead of five hundred. Pinterest’s discovery works because the algorithm picked. Spotify’s Discover Weekly works because it’s 30 songs, not 3,000. Letterboxd recommendations work because the social signal narrowed the field before you got there.¶

The pattern is the same. Someone, or something, did the work of choosing.¶

This shifts what design has to optimize. Less how do I rank better and more how do I narrow gracefully. Less show more options and more show fewer, better. The interface has to invite the clarifying question, not bury it under a search bar that pretends to be a feed.¶

Most of the AI products I’ve used in the last year fail this test. The ones I keep using pass it.¶

I’m building in a category where this matters acutely, and I see the pull on every team I talk to. Show more inventory. Run a wider search. The metrics dashboard rewards throughput, and the interface drifts toward the thing the dashboard rewards. The product gets bigger when it should get sharper.¶

The flip is hard because it requires conviction about a number going down. Fewer results per query. Fewer paths off the home screen. Fewer feature buttons on the surface. The product gets smaller before it gets sharper. That’s a hard story to tell a board.¶

But the products people love in this space already made that flip. The Wirecutter doesn’t try to be Amazon — it picks one toaster and tells you why. The same shift is happening to design teams — fewer people, sharper output. Across categories, the move is the same. Fewer suggestions, better ones.¶

The leverage isn’t in producing more. It’s in choosing well, on behalf of someone who didn’t have time to choose for themselves.¶

Ask better. Bring less. The five right things beat the five hundred almost-right ones every time.¶