A mere 2x productivity improvement sounds like something you could achieve by introducing new tools that are predictable (i.e.: Not AI).
It's strange that the goalpost seems to have moved from "AI is net negative to productivity" to "only 2x improvement isn't worth it"