HBR is analyzing this with an old world lens. It might very well be that the effects are as they say temporarily. But the reason this is happening is because AI is in fact replacing human labor and the puppeteers are trying to remain employed. The steady state outcome is human replacement, which means AI does in fact reduce human labor, even if the remaining humans in the loop are more overloaded. The equation is not workload per capita but how many humans it takes to accomplish a goal.
If AI was indeed replacing human labor I would expect HBR to be among the first publications to cover it.