I came in with a long list of features. A map. Filters. Tags. Saved places. Friends. Reviews. A whole little social layer. I was sure all of it mattered. Bwendi politely listened to all of it and then asked one question that I have not stopped thinking about: what is the smallest version of this that still answers the only question a hungry person actually has?
The only question is “where, near me, right now, is good and open and honest.” That is one question. Every feature I had been excited about was a different question. The first thing they did was throw most of my list away on my behalf. They were nice about it. They were also right. The product is good now because they were willing to do that on day one instead of six months in.
They modeled the user, not the database
The biggest difference between the early sketches I had and what Bwendi shipped is whose head the screen lives in. My early version was organized like a database: categories, tags, distances, ratings, all equally weighted, all visible. Their version is organized like a person standing on a corner at 7pm who is hungry. That sounds obvious. It is not. Most apps in this space are still built around the data shape, and you can feel it the second you open them.
They did this by, at every step, asking what state the actual human was in when they would touch the screen. Hungry. Tired. In a place they don’t know. Holding the phone in one hand. Sometimes a little drunk. The interface they designed reflects that. The decision is short. The signal is loud. The chrome is almost invisible. You can use the thing without reading it.
Place context as a first-class object
The one technical idea I am the proudest of, and that I cannot take credit for, is how they treated place context as a first-class thing in the model rather than a filter. Time of day, day of week, neighborhood rhythm, whether the regulars are inside, whether the kitchen is at the wall it should be at — in their model, those are not garnish on top of a restaurant record. They are the record. A restaurant means something different at 1pm Tuesday than it means at 11pm Saturday, and the model respects that.
That is also why the product can give you a confident, narrow answer instead of a long ranked list. If the model already knows what kind of moment you are in, it doesn’t have to hand you twenty options and ask you to do the work. It can hand you the right one. That confidence is not magic; it is the shape of how Bwendi modeled the problem.
They removed things until the product got obvious
I cannot count how many times in the build I’d open a prototype and find that something I had insisted on the week before was just gone. There would be a short, friendly note explaining why. Almost every time, the screen was better without the thing I had insisted on. After a while I stopped fighting it. They were doing the editorial work I should have been doing on myself.
The whole product has the same restraint. There is no streak. There is no badge. There is no friend graph nudging you to share your dinner. There is no “you might also like.” Each of those was a real conversation we had and decided against. The reason the app feels like a tool and not an app is that someone, repeatedly, on my behalf, said no.
The reason it feels like a tool and not an app is that someone, repeatedly, on my behalf, said no.
They prototyped with real places, not lorem ipsum
From the first interactive sketch, the screens were full of real restaurants in real cities at real addresses, with real opening hours and real awkward edge cases — the place that closes for two hours in the afternoon, the place that is technically open but only serves drinks until 8, the place with a different name on the door than on the internet. The decisions they made about layout and hierarchy were stress-tested against actual messy reality from day one.
That is why corner cases that would normally only surface after launch were already designed for. The app does not break when reality is awkward. Reality is almost always awkward. They built the product as if they already knew that.
They respected our voice
The other thing I want on record is that they let the writing be the writing. None of these field notes are the kind of corporate house voice that gets bolted onto products of this size. Bwendi pushed for the opposite — that the voice should be specific, opinionated, and clearly mine, because that is how you build trust with someone deciding where to eat tonight. The site you are reading right now is part of that argument made physical. They built the rooms. I get to talk in them.
What I’d tell anyone thinking of working with them
Come with the problem, not the solution. They will turn your problem into a much smaller problem than you thought it was, and then they will solve that smaller problem properly. You will leave with fewer features than you came in with and a much better product. If you are trying to ship something honest quickly, with no extra weight, that is the trade you want.
HungryButNotStupid exists in the shape it does because of this. The field notes you have been reading, the design of this site, and the product they point at are all the same conversation, and Bwendi is on the other side of all of it.