van Inwagen pushed this argument as a libertarian. So, the below argument isn't prejudiced by compatibilist assumptions.
I want to close by explaining why van Inwagen thinks one important group of incompatibilists, those who appeal to what is called agent-causation, do not appreciate the depth and difficulty of the problem of free will. Many philosophers would agree with this judgment for the simple reason that they think that the concept of agent-causation is incoherent, or think that agent-causation is metaphysically impossible. Van Inwagen is inclined to agree with them (although he has no firm opinion on this question), but he has lately stressed a different point. It is this: suppose there is nothing conceptually or metaphysically impossible about agent-causation; suppose in fact that agent-causation is a real phenomenon and that an episode of agent-causation figures among the antecedents of every voluntary movement of a human hand or limb or vocal apparatus. Van Inwagen’s position is that even if this is so, and even if (as some have argued) we understand the concept of agent-causation at least as well as we understand the concept of event-causation, all this does nothing to diminish the mystery of free will. I will try to explain why van Inwagen thinks this by considering a particular human action. Suppose Marie wants to vote in favor of the proposal before the meeting, and that, for this reason, she raises her right hand when the chair says, “All in favor. . .?” Suppose that one of the causal antecedents of her hand’s rising was a certain event in her brain that was undetermined by past events, that the state of her body and her immediate environment at the moment this brain-event occurred was causally sufficient for her hand’s rising, that if this event had not occurred, her hand would not have risen, and that she, Marie, a particular member of the metaphysical category “substance” or “continuant,” was the cause—that is to say, the agent-cause—of that crucial brain-event. The friends of agent-causation, if van Inwagen understands them, believe that these suppositions are sufficient for her having freely raised her hand. If that is so, these suppositions must entail the following proposition: at some moment shortly before Marie raised her hand, she was able to raise her hand and she was able not to raise her hand. But van Inwagen doesn’t see why this entailment should be supposed to hold. In fact, he thinks he sees a good argument for the conclusion that it was not up to her whether her hand rose. Suppose God were miraculously to return the world to precisely the state it was in, say, one minute before Marie raised her hand, and that he then allowed affairs once more to proceed, without any further miracles. What would happen? What would Marie do? Well, if her raising her hand was a free act, and if free will is incompatible with determinism, then we can’t say. We can say only that she might have raised her hand and might not have raised her hand. If God were to cause this episode to be thus “replayed” a very large number of times, it might turn out that she raised her hand in thirty percent of the replays and refrained from raising it in seventy percent of the replays. This much is a simple consequence of incompatibilism, and it brings one of the main reason philosophers become compatibilists into stark relief. It seems to lead us inescapably to the conclusion that on each particular replay, what Marie does on that occasion is a mere matter of chance. And if there are no replays, if there is only one occasion on which Marie is in this situation, it seems to lead us just as inescapably to the conclusion that on that one occasion what Marie does is a mere matter of chance. And if it is a mere matter of chance whether Marie raised her hand, then it cannot have been true beforehand that Marie was both able to raise her hand and able to refrain from raising her hand, for to have both these abilities would be to be able to determine the outcome of a process whose outcome is due to chance. It is true that we have, by stipulation, inserted into this process, this process whose outcome is due to chance, an episode of agent-causation. But, if I may so express myself, so what? That doesn’t change the fact that the outcome of that process was due to chance. If God caused Marie’s decision to be replayed a very large number of times, sometimes (in thirty percent of the replays, let us say) Marie would have agent-caused the crucial brain event and sometimes (in seventy percent of the replays, let us say) she would not have. Surely, then, whether she agent-caused the brain-event was a mere matter of chance? Whether her deliberations were followed by her agent-causing the brain event was, it would seem, a matter of chance; Marie, therefore, cannot have been both able to agent-cause the brain-event and able to refrain from agent-causing the brain-event, for to have both these abilities would be to be able to determine the outcome of a process whose outcome was due to chance—an impossible ability. I conclude that even if an episode of agent-causation is among the causal antecedents of every voluntary human action, these episodes do nothing to undermine the prima facie impossibility of an undetermined free act. Postulating agent-causation, therefore, does nothing to diminish the mystery of free will. Van Inwagen’s conclusion is that incompatibilists had better abandon the concept of agent-causation, and seek a resolution of the mystery of free will elsewhere—if, indeed, there is an “elsewhere.”"