AI automation amplifies demand for expert human judgment rather than replacing it

The data annotation industry reveals a counterintuitive pattern: as AI gets smarter, the demand for humans is rising, not falling. Pre-labeling can slash costs by 100,000x for simple bounding boxes, and Active Learning loops cut labeling effort by 35% while maintaining identical accuracy. But the result is not fewer humans — it is fewer humans doing higher-value work. Projects that needed 500 contributors now need 100, but expert rates for medical diagnosis or legal annotation hit $200/hour.

This directly extends The intelligence-to-judgement ratio determines which professions AI automates first: intelligence work (simple tagging) automates first, while judgment work (expert RLHF, red teaming) remains human and commands premium compensation. The pattern also connects to Tribal knowledge is the irreducible human input that enables agent automation — the most valuable annotation is domain-specific expert judgment that no automated scan discovers. Nadella generalizes this to the whole firm in Every firm must build two capitals — human capital and token capital — and they compound together, not at each other's expense: human capital becomes more valuable, not less, as AI capability grows, because human agency is what directs the compute.