This started as a podcast club at Gesamtschule Barmen, running through the 2025/26 school year with students from grades 9 and 11, led together with a teacher. The plan was to do one real project and learn on it: how to ask someone for something, how to write an email, how to use AI, how to take a thing from start to finish and keep track of where you are. senseBoxes, small open measuring stations, turned out to be a good project for that.

The data had a purpose. Some rooms at the school get too warm, and measurements make that harder to ignore. The group settled on the common area, with its huge glass front and a lot of plants. Once the spot was picked, there was nothing there to fix a box to. So we designed holders together and 3D-printed them. In the end we only got one station up.

Then the data sat in a database nobody looked at, and the next problem was software. We wrote it with AI, together, and most of the work happened before the first prompt: what do we actually want, what has to happen when the network drops, what does someone in the hallway need to see. The AI is fast at writing code and bad at guessing what you meant, so the requirements went on paper first.

The first piece is an extension for home automation. It pulls any open station in the area into your own system, where its readings behave like your own sensors and land in the long-term statistics. If a station goes quiet for ten minutes, it raises an alert. Otherwise a station that belongs to nobody fails and nobody notices.

The second is a dashboard for a TV: a screen that runs fullscreen, fetches whatever the station measures every minute, and puts a traffic light on every value. You don’t have to read a number to see that it’s time to open a window. Nothing flickers on an update, and when the network drops, the last values just stay up. A screen in the hallway that blinks every minute gets switched off.

At the end the club went back to what it started as. The students made a podcast about how it went: what they built, what went wrong, what they learned. It was only played at school.

None of these steps was planned at the start. Each one solved a problem and turned up the next, and the skills came at the point they were needed. Design research calls this the co-evolution of problem and solution (Dorst & Cross, 2001): the problem only takes shape through attempts to solve it. Janet Kolodner’s Learning by Design (2003) builds school lessons on exactly that, with the design task first and the science learned along the way. Seymour Papert’s constructionism adds why it sticks: you learn best by making something other people can see, a holder on the wall or a podcast someone listens to.