One useful step in developing Koda, my reading and writing project, was to write down what happened while I was using it.
I saved a short note inside the prototype. Some tool explanations were difficult to understand. Opening the mentor took me away from the writing page. After opening the export menu, it stayed open when I wanted to continue working.
These observations gave me something more concrete than “make the interface simpler”.
I asked for the mentor to open alongside the writing, for tools to be arranged more clearly, and for the export menu to close when I moved on to another action. I also asked for unnecessary explanations to be reduced.
AI helped organise the notes into a change list and supported implementation. This was an iteration based on my own use of a local prototype. It did not establish that the changes improved writing quality or worked for a wider group of users.
What I want to carry forward is the structure of the feedback: the action I was trying to take, what interrupted it, and the behaviour I expected instead. That gives a product requirement a practical starting point and makes the next check more specific.