It seems to me that the author fails to extrapolate on the effects of recursive self improvement. The only things preventing 95% engineer obsolescence will be compute/energy constraints and the speed of adoption, which can take years for large infrastructure companies. But it's coming.