I spent hundreds of hours thinking about boundaries and ran two related AI safety workshops. All because every explanation of boundaries I had ever heard contradicted itself.