If one walks down to the station and confesses to having planned and executed a killing and requests an attorney; I’d be inclined err on the side of ‘probably a moral agent’, barring evidence to the effect that it was just scripted to do that.
This question is deeply nontrivial because we don’t really have anything approaching clear boundaries for agency in humans(it is more or less agreed, at least for operational purposes, that humans are agents; but push any of the boundaries and it devolves into vague hand-waving pretty quickly: how old is old enough to be an ‘adult’ and thus a full agent? How drunk is too drunk to be considered to be exercising agency? What, if any, dumb mistakes are caused by bounded rationality that the person should be protected from, rather than merely indicating economically irrational preferences? etc, etc.); but I don’t see any reason why sexbots would be considered any differently than SCADA systems, aliens, or rats of NIHM for the application of “Well, did it kill somebody and is it a moral agent?”
