Are we overestimating how quickly AI capability turns into real productivity?

​ I’m not questioning whether AI models are powerful. They clearly are. But I’m starting to question whether people underestimate the distance between “capability” and “productivity.” A model can produce a good answer. But productivity in the real world often requires persistent context, judgment, tool access, process knowledge, responsibility, and integration into messy human systems. This seems especially important in the AGI discussion.