It is for families. You collect recipes from anywhere: Instagram, YouTube, a blog, or a photo of a cookbook page. They land in one cookbook everyone in the house can add to. Then you plan the week together, and the shopping list writes itself. Three of us build it: Firuze, David and me.
Between work and small children there is no room to think of something, shop for it and cook it. Our recipes lived on Instagram, in browser bookmarks and in a cookbook on the shelf, never anywhere you could actually plan from.
Every app did one piece of it. A recipe box, or a meal planner, or a shopping list. Never the whole way from “that looks good” to “it is in the basket”. And almost nobody cooks just for themselves. It is for the family, or for friends coming over, and that was the part nothing else seemed to notice.
The whole film is a script, not an edit suite. The app footage is the real app, recorded in the iOS simulator against a demo household. The people and the kitchens are generated: Kling for the faces, FLUX and Gemini for the rooms. The voice is Gemini TTS, the motion graphics are Remotion, the cut is ffmpeg, and the music is synthesised in code because there was no music model on the gateway to ask.
Every model runs through one gateway, so the film is a command and a storyboard file. I wrote the pipeline with Claude Code and then re-cut it about twenty times, which is the part that took the days.
Every piece I needed I had to learn first and then hand write. Somewhere in the middle the idea ran out of road. It never shipped.
I used to shoot weddings, so the photographs in it were properly good. Then almost nobody else added anything. People told me afterwards that their phone snaps looked shabby next to mine, so they simply did not post.
Same idea, same person, the better part of a decade apart. Implementation stopped being the bottleneck, so the work became deciding what the thing should be, and that part I had already done.
The AI studio was in it from the first weeks rather than bolted on later. Photograph the bowl badly, in bad light, with one hand. Nobody has to hide, so everybody adds.
Parental leave, a newborn, and no desk to sit at. A Mac Mini stayed in the office running a set of agents, and I described what I wanted from my phone and kept iterating. On the sofa, on a walk, in the twenty minutes a baby hands you. I wrote the whole thing down, including the review I skipped and the builds that shipped broken.
And almost every evening I cooked from it. The shopping list on the phone in the supermarket, the recipe propped on the iPad by the stove, importing whatever Firuze had found on Instagram that day. Build it in the morning, cook from it that night, fix what annoyed me the next morning. Very few products get a loop that short, and none of my earlier ones did.
It had to live where households already are: a link in a group chat, a photo of a page from a cookbook, a plan the other person can open without installing anything. That ruled out an app first build. It also ruled out a website, because timers, widgets and notifications are the reason anyone opens a cooking app at all. So I built both, out of one codebase, and spent the engineering on the seam between them.
Most of the recipes people actually want live in a reel, and a good share of those posts have no written recipe at all, just someone cooking while they talk. Every competitor handles this the same way: it works when there is a caption to parse, and when there is not, it quietly fails. Read the reviews of any of them and the same comment comes back. Tried it, a few videos did not work, gave up.
So I built the expensive version. The Mac Mini pulls the video down and a model watches it, actually watches it, and describes what the person is doing and saying. That becomes the recipe. It costs more per import and it takes longer. It also means the answer is never “that one did not work”.
Single biggest jump in quality the product has had. Nobody asks for it by name. It is the reason people stay.
A few households cook from it while we build. These are their messages, in the chat where they sent them.
We always have the same problem. The problem is time to think about it, get the ingredients and do it. With 3 kids and a puppy, time vanishes.
You totally nailed the problem.
I used the photo import for the pancakes. Worked well!
Wow, the recipes are seriously amazing.
Wooow! Awesome! I was waiting for it.
Still a closed beta, and not in the App Store. The app reads a reel and hands back a recipe, which only helps if you believe the recipe. A wrong quantity costs more trust than a good import earns, and a household gives you a few chances before going back to the browser bookmark it already had.
The design, the code, the App Store listing, the support replies and the marketing that got anyone to look. If you have an idea and the means to back it, that is the whole offer: I can carry it from a conversation to a thing people use.
Smaller doors too: coaching someone building their own product, or taking yours apart with you and handing over how I solved the part you are stuck on.
