Jeremy Somsouk

I couldn’t sleep, so naturally I started writing specifications for an app.

That wasn’t quite the plan for the night. But I had been thinking about birth wishlists, looking at Milirose, and getting frustrated with how messy the whole experience felt. I kept thinking about what I would want to use myself. Something easier to navigate, easier to organise, and comfortable to use on a phone.

That was the beginning of Bientôt.

Something I would want to use

Starting with my own needs made it easier to find ideas. I didn’t have to invent a hypothetical user and guess what might bother them. I could look at what existed, try to do something, and notice where I got annoyed.

There’s a limit to that, of course. What makes sense to me won’t necessarily make sense to another family. But it gave me somewhere concrete to start.

The wishlist was the first thing. Then I started thinking about the preparation around it. Choosing what you need, keeping track of things, sharing that with someone else. There seemed to be room for something useful there.

And I kept coming back to the phone. If I’m checking something quickly or sharing a list with family, that’s probably what I have in my hand. I wanted the experience to feel like it had been built for that from the beginning. Some of the existing websites made me feel like I should come back later with a computer.

Two hours of specs

Once I started writing, I kept finding things to add.

What should the app do? How should it behave? What would I expect when I came back to it? Writing one answer usually produced another question.

About two hours went by like that. Sleep still wasn’t happening, but at least I had a growing specification.

I continued the next day. By then, I was also curious about a second part of the project: how far could I get by giving a coding agent as much clear context as possible?

I wanted to put the effort into describing what I wanted, then see what a relatively inexpensive model could achieve with it. I picked GPT-6.1 Sol. The cost and efficiency were part of the experiment. I wanted to see how far I could go with a model I could afford to keep working with through all the iterations.

I knew there would be iterations. I just had very little idea how much we would manage to build.

I expected less

I’m an engineer, so I wasn’t going into this without knowing what sits behind an app. The screens are one part. There’s also the architecture, the data, the configuration, the deployments, and all the small things that need to agree before anything works.

Still, I didn’t expect to get this far.

The surprising part was how much of that whole process I could work through with the agent. We could discuss the architecture, implement it, work on the design, then deal with deployment and configuration. The project kept moving into areas I had expected to become stopping points.

That didn’t mean I could write a specification, disappear, and come back to a finished product. There were decisions to make, things to check, and results that needed another pass. Sometimes I knew exactly what bothered me. Sometimes I had to use the app before I could explain it.

But I could try something, see it working, and decide what to change. That made it very easy to keep going.

And then it became a project

Bientôt started with a fairly ordinary frustration: I wanted a better wishlist experience for myself.

During a night without sleep, that turned into a pile of specifications. Then it became an experiment with a coding agent. Somewhere along the way, I started to believe it could become useful to other families too.

I don’t want to tidy that up into a grand founding story. I was annoyed with a website, I couldn’t sleep, and I got carried away writing down what I would prefer.

The next day, I still wanted to build it.

So I did.