Working around AI and data infrastructure makes the questions more concrete: what is reliable, what scales, what creates leverage, and what should remain human?