That's still a lot of state. How do you know an object is in the right state to perform its behavior? Well, acting as some other object you have to either read the first objdct's state or model it internally. And now we're coupled on state and behavior.

So what if the first object ensures that it's always in the right state to perform it's behavior? Well, now it's observationally equivalent to a state-free actor. Why did you involve state in the first place?

Because languages make it hard to not do so?