Yes, it varies a lot. But by and large cases where robots wildly outperform humans were automated decades ago, because the obvious benefits justified the cost and complexity of either building a bespoke robot or programming one to do the job well (all those robot arms you imagine swinging doors into place at the car factory)

The cases left over (and discussed here) are either requiring a level of flexibility that older designs of robot couldn’t handle, or where humans were pretty efficient at anyway, so the complex process of prepping a robot wasn’t justified.

But a robot that can be taught without programming (by any worker or their supervisor), and slots straight into an existing human-shaped hole? That could massively reduce the upfront cost, especially if economies of scale make the robot itself cheaper. possibly to the point that the robot could be worse at the job than a human and still be cheaper in the long run.

It doesn't have the best robot for the job, it only needs to be more profitable than a human for that particular job, and be a "slot in replacement" with low risk. The free market system is very inefficient that way lol.