I spent hundreds of hours thinking about boundaries, spoken to 100+ people, and run two related AI safety workshops. All because every time I asked someone to explain boundaries, their explanation contradicted itself.