Cheap, abundant AI-generated text raises the marginal value of rare, high-signal human writing, which implies the scarce resource shifts from generation to discrimination — from producing content to identifying which content is distinctive.
Why it matters
If generation is commoditized, then engineering effort should shift toward provenance, uniqueness scoring, and curation rather than raw throughput; evaluation sets and training corpora become the bottleneck asset. (This is an inference from the speaker's claim, not a demonstrated result.)
Generalization
In any domain where a model collapses the cost of producing outputs, the value migrates to the selector, the verifier, and the rarity detector — not the producer.
AI produces prodigious amounts of new text and writing, potentially elevating the value of extraordinary human writing.