The first point has an obvious response and the last is speculative.
I find interesting to address the middle one statistically.
A given machine has a number of random events that can kill it. Each possible event type i can be modeled as a Poisson process Xᵢ with, in a given year, probability λᵢ. The machine survives in a given year if no occurrence of event i happens, with probability Pr(Xᵢ = 0) = exp(-λᵢ).
For instance, a device with an average lifetime of 8 years and a single possible cause of death, has λ = -ln(½^⅛) = 0.087.
Given N independent event types (say, if you have on the order of N components), the yearly survival of the overall machine has probability Πᷡᵢ exp(-λᵢ).
Thus the lifetime of the overall machine is ln(½)÷ln(½^(Σᷡᵢ 1÷Lᵢ)) = 1 ÷ Σᷡᵢ (1÷Lᵢ)
if each component has lifetime Lᵢ.
For instance, two components with an average lifetime of 8 years yield a single machine with an average lifetime of 4 years.
The more death-causing event types there are, the smaller the lifetime; and a general-purpose robot inherently has more.
I believe this result is counter-intuitive, and is the reason you might be led to think I am overly pessimistic.