I Wasn't Building an App. I Was Training a New Way to See Food.

For a while, the sharpest hours of my day were spent arguing with a machine about a carrot.

I'm not being cute. The screen would be lit at some indefensible hour, and I'd be going back and forth with an AI over one question: when Foodna sees a carrot, what should it actually see? "A serving of vegetables"? Too coarse. "Beta-carotene"? Still not enough — because a raw carrot eaten whole and a carrot chopped and softened in a little olive oil are not the same sentence to the body. In the first, most of those fat-soluble carotenoids pass straight through, untouched. In the second — cell walls cut open, loosened by gentle heat, with fat riding alongside — the body actually catches them. Same carrot, two fates. What I wanted the system to see was the fate, not the name.

That night, we went around on that one carrot for a long time.

I tell you this because it marks, precisely, the thing that happened to me over these two years — and that thing is far larger, and far stranger, than "I made an app."

I used to understand exactly how a product gets pushed into the world. It takes a group of people: someone holding design, someone holding engineering, someone guarding the data, someone in the middle grinding all those judgments into alignment. Inside a large organization, what you learn is how to get many hands pushing the same direction at once. It's a real skill; I won't pretend it isn't. But it also means the most private, least-willing-to-compromise standard in your head has to pass through layer after layer of translation, dilution, negotiation before it ever touches the product — and by the time it lands, it's usually no longer the shape you first wanted.

Foodna was the first time I held every boundary myself, alone, start to finish.

What made that possible was AI. But let me be exact about how it was possible, because this is the part that gets told wrong.

People love to say AI makes you faster. True — but that isn't the part that mattered most to me. What actually stopped me cold was something else: AI, with no mercy at all, put a truth in front of me — if the founder has no taste, AI just produces mediocrity fast. It can hand you ten options in seconds, each one fluent, each one reasonable, each one sounding healthy and sensible. And that fluency is exactly the dangerous thing. It manufactures the mediocre quickly and prettily — quickly enough that you almost don't have time to ask: is this really the one I wanted, or just an answer that sounds like it couldn't be wrong?

So the thing I did most over these two years wasn't generate. It was refuse. Sitting inside a pile of things that all looked correct and saying, again and again: not this one. Again. This one is too much like what a nutrition app would do. This one is too much like a medical dashboard. This one is too much like a supplement brand's patter. None of it is what I want. I want Foodna to see food my way. And slowly I understood: I wasn't using AI to build a thing. I was using AI to take an aesthetic and a standard that existed only inside my head, and train it, for the first time, into a system that answered back to me.

In all this, the AI played many parts at once. It was my researcher, helping me map the edges of one paper after another. It was my design assistant, sketching my vague intuitions into something concrete enough to reject. It was my engineering partner, turning ideas into things that ran. It was my best sparring partner — especially on nutrition, where I needed something to keep contradicting me, forcing "I think" into "I think, with grounds." It was even my first reviewer. But there was one thing it never did, and shouldn't: the final judgment. Which signal actually matters, which sentence counts as honest, what a carrot should be understood to be — that was always mine.

I want to put this line down precisely, because it's the center of everything: AI didn't lower my standards. It just made me able, for the first time, to hold a standard this high all by myself. I used to have to convince a roomful of people to move the standard an inch. Now I don't have to convince anyone. I only have to refuse, over and over, to settle.

And that is why Foodna grew into what it is — and what it grew into is the thing I'm most stubborn about.

It is not a calorie counter. Not a cold medical dashboard. Not a supplement brand. Not one more fitness check-in app. I wanted to build none of those. The reference living in my head is actually a long way from "tech product." It's closer to the menu of a very good restaurant — few words, every word weighed; a private food magazine made for an audience of one; an ingredient archive kept with real ceremony. It has the restraint of omakase — not more, only the right thing. The honesty of Nordic materials — no ornament, letting a thing look like itself. And the color of genuine Mediterranean ingredients — the red of a tomato, the green-gold of olive oil, the almost-luminous purple of red cabbage.

I hold, stubbornly, to one belief: screens are making real food less and less wanted. A filtered dessert, a takeout dish plated for the algorithm, will always out-shout a plain plate of sardines and lemon on a screen. So if I want people to want real food again, I have to make real food worth wanting on a screen. Beauty here is not packaging. It's a reward interface — it has to make a healthy, restrained, real meal still land, still catch you, in the half-second you swipe past it. No one else could vet this, because it's too subjective — subjective enough that only one person's taste gets to decide.

I always assumed that holding a standard this fine, this personal, would be a lonely thing.

And then the outside world started coming in.

Friends first. They tried the demo while it was still rough, and what they came back and said wasn't politeness. It was "I'd actually keep using this." It was "this is the first time logging food didn't feel like being judged." Then people I'd never met — in Madrid, in San Francisco — people as exacting about food and their own bodies as I am. Watching them react to Foodna, feeling that sudden, warm, direct hit of recognition, I sat still for a long time.

I sat still because I had believed this exactness was a private quirk of mine. I thought I was the only person on earth who'd lose sleep over how a single carrot should be seen. I wasn't. It turns out there's a whole group who have been waiting for a smarter, more beautiful, less shaming health tool — one that doesn't tell you you went over today, but tells you you just gave your body a good set of things.

That was when I finally understood: what I'd made over these two years was never just an app.

I was training a way of seeing food — training the machine first, and, really, training myself, taking that intuition that had always lived vaguely in my body and making it, piece by piece, clear, reliable, shareable. It's a form of creation I had never experienced before: taking something intensely personal — aesthetic, science, the body's own experience — and, alone, training it into a system that responds to you. It no longer lives only in my head.

Foodna isn't AI seeing food. It's me, using AI, training a way of seeing that comes closer to the one I always wanted. And now I know I'm not the only one who wants to see food this way.