Ultimately one man can do only so much. Thus you start to involve middlemen who do not necessarily make optimal decisions - and you have to predict their result

Ultimately one man can do only so much. Thus you start to involve middlemen who do not necessarily make optimal decisions - and you have to predict their result at least in the median. Stack many levels of this, each accumulating error and delay. Good luck. There is such a thing as control efficiency gain, which you have to balance against stability and constraints. It's a really tough problem, and since economists only work with slow averages, the gains they can make are relatively small over the general growth if present. And that presumes your optimization condition is ultimately correct and complete, which most of these are not. Even an oracle would have limited efficiency unless it could reproduce itself. Even if you can observe all states that does not mean you can control them.