Every August in Sweden there is a moment where everyone around the table suddenly knows a song and you do not. It is crayfish season, the little red things are piled up, someone lifts a shot of something strong, and the whole party breaks into a snapsvisa you have never heard in your life. The serious parties even print song books for the occasion.
Our family kraftskiva goes a little differently. The only songs all of us actually know, kids included, are kids songs. So our snapsvisor have been Let It Go and Baby Shark, sung with total commitment before the grownups down their drink. It works. But it did make me wonder what it would take to walk into a party knowing the songs for once.
I had spent a while building small apps with AI helping me write the code. This time I wanted the opposite: an app with AI actually working inside it, generating something for the people using it. Crayfish season was coming up. The two ideas met.
An RSVP that grew a second half
As usual, it started out as an event sourced RSVP app. You make a party, people say whether they are coming, done. That part was quick.
Then, right as I finished the RSVP, the better idea showed up. What if guests could also answer a few questions when they replied? And what if the host could take all those answers and, with one tap, turn them into a set of songs for the party?
That is the app now. Guests RSVP, including the very Swedish “kanske”, maybe, which still counts. They answer a handful of questions, and once two or more people have answered the same one, it can become a track. The host picks which tunes they think everyone at the table will actually know, the melodies people can sing without a songbook. Then the app writes fresh snapsvisor to those tunes, using the answers as raw material. Songs about your specific party, set to melodies your specific guests already have in their heads.
A record, pressed for your party
The nicest accident was the name. In Swedish a skiva is a party, and a skiva is also a record, the disc you put on. So I let the whole app lean into that.
A generated set of songs is a pressing. Your collection of them lives on a shelf called the Skivhyllan. When one is being made, a little crayfish spins like a vinyl on the dashboard while it presses. And any single track can be shared out as its own lyric card, with a small “written by” credit for whichever model wrote it.
My favourite part is what happens when one more guest replies after you have already pressed a record. It does not throw the record away and start over. It just splices that new person’s verse into the song and keeps the rest as it was.
Three models walked into a party
The generation was the whole point of the exercise, so I did not pick a model on faith. I ran the same job through Claude, Gemini, and OpenAI and read the results side by side.
I have to say, Gemini was the clear winner when it comes to creative writing. When the task is silly and warm and a little bit clever, it just had a feel for it the other two did not. Which is the kind of thing you only find out by actually putting all three to work on the same prompt.
Limits, for free, because of course it is event sourced
The first version let you regenerate as many times as you liked, but I thought that might easily drain my budget, since every press is an API call that costs outside my normal AI subscription. So it was a good exercise in putting limits in place: three free presses, and campaign codes to unlock more. And that was easy because, you guessed it, the whole thing is event sourced. A press is just an event on the pile, and your allowance is three plus whatever codes you redeemed minus what you have used.
New features like these were pretty easy to add on without having to muck around with any database. Maybes that still count, questions that need two people, tune pools you can edit after the party is made, the paywall. Just new facts, and projections that fold them in.
So, which one is most creative?
Here are the same party’s songs, written by all three models. Same tune, same guest answers, three very different records.
Which do you think is the most creative? I have my pick, but I am curious whether you land on the same one.
Anyway. Kraftskiva season is almost here, and this time I might even know the words.
being in tech 