In the face of perfect, self-modifying robot laborers, YES. All human labor will become less efficient, economically, than using the machines. This includes top level positions, like CEOs, because the AIs will be better at forecasting and acting on very complicated long term strategies, with deep levels of nested contingencies-- and as more and more actors become AIs, the ability of the actors to predict each other will increase.
Make them emotionless, and you have robot janitors who have no emotional distress at being janitors, and the robot CEO values the robot janitor and its contribution to the operation of the robot-run business consistent with the calculable value (in terms of upkeep, prevention of malfunctions in ambulating robot workers due to an unobstructed path for movement, etc) that said robot janitor provides.
In short, humans, and the whole profit system they have created, will be wholly superseded by robots, if perfect self-modifying AI is ever developed.
For the same reason that employers do not want to train people but instead want "Perfect fit" employees, in this post-labor environment, there is simply no benefit to hiring humans to do--- ANYTHING. The robots with advanced AI would simply be better at everything, require fewer resources, can work longer hours-- even endlessly, barring maintenance-- don't complain about work conditions-- in fact, dont even require breathable air--- so cheaper to run the factories anyway simply from environmental needs-- etc.
The disadvantages and cost centers associated with employing humans would make any business that does so less fit in that new market, and their competitors would quickly overtake them.
Robots do not require "creativity", if they have comprehensive understanding of the problem domain. They can computationally derive the absolute maximum possible efficiency, and then shoot directly for that. Human creativity is the result of biological computation's limits being resolved in a "good enough is good enough" manner via evolution. real, strong general AI could shoot for absolute ideal, and achieve it. Any product designed by robots, would ultimately become better than anything humans could design, given that the robots fully comprehend the problem domain involved.
Take for instance, the problem domain of a game of Go. All possible legal moves for this game have now been computationally derived.
http://tromp.github.io/go/legal.htmlThis means that a general AI could make use of this solution space, and with it (and with enough processing power) outmatch any human Go player.
Machines dominate by brute force through knowing the problem domain. They do not need creativity.