Chapter 11 of the 2008 version of On What Matters introduces a topic not touched on in comparable sections of Climbing the Mountain or the 2002 Tanner Lectures, the topic, namely, of free will and desert. In this chapter Parfit discusses the question of the sorts of compatibilism Kant considers and what is needed for morality.
Parfit opens the chapter by mentioning the transcendental distinction Kant makes between noumena and phenomena and essentially states a version of the Groundwork III argument that noumenal freedom is what underpins morality (though there is also a version of this argument in the Third Antinomy). The point of the argument, though Parfit's discussion is not textually grounded, is to provide a ground for taking morality not to be an illusion. Parfit does accept the general "ought implies can" move Kant makes, though, notably, in making this premise central, he is following a version of the argument Kant makes that is quite different to the Groundwork III discussion that he appears to begin with.
Parfit does not, however, take the general argument he reconstructs from Kant to be persuasive, primarily because of the way Parfit views the notion of what it means to say that one "could" have done something different to that which one did which Parfit views only in a hypothetical sense, although, in saying this, he seems not to notice he is repeating an argument from Leibniz that Kant would precisely not be happy with. After all, the guarantee of merely "hypothetical" conditionals proves nothing about the reality of freedom.
The reason why Parfit takes this shift to be sufficient is because he thinks Kant confused determinism with fatalism and that the fatalist view is the only one that implies that it makes no difference what we do. However, this is not the point at all. The determinist view, as Kant construes it, allows for universal causal reasons to apply for all actions and these reasons are both necessary and sufficient to account for what takes place. Given this, the addition of "hypothetical" conditionals makes no difference to the reality of what has taken place which is why Kant describes it as a "wretched subterfuge".
Parfit essentially approaches determinism in a practical rather than a speculative way insisting on the point that there is nothing to prevent us from acting merely because there are reasons why what is taking place is happening as it is. The essential point Kant is making, however, concerns the origin of the basis of my action, that is, whether it arises only externally to me or whether there is some internal, spontaneous, ground for action. Questions of hypothetical conditionality do not help resolve this question.
Having failed to "correct" Kant in this way Parfit moves on to the analytically distinct question of what is involved in claiming that someone "deserves" to suffer and he focuses on the Kantian claim that if acts were merely events in time (hence all fell under a universal causal claim) then we couldn't be said to act in ways that would produce suffering as desert. Parfit seems, by contrast, not to think this claim about desert is required at all for moral statements to have sense. One of his reasons seems to be the rather weak one that either events are causally determined or they are partly random. The reason Parfit takes this disjunctive to be sufficient is that reasons given will all end somewhere and thus be partly random. However this is simply not the case. Reasons end in a final appeal to a basic principle taken to be sufficient which does not make actions partly random but rather shows a basic divide between reasons for action that has to be overcome (though this may be, as Kant argues in the Religion, by means of moral "conversion").
Parfit goes on to look at Kant's claims about moral character and arrives at a better view when he returns to his earlier claim that it rests ultimately upon the noumenal freedom thesis. However Parfit proceeds to reject this thesis as he does not find it intelligible. What Parfit seems to be mean by this is that a theoretical defence cannot be elaborately made of it though this is not even required by Kant since all that he sets out is that the case against freedom cannot be definitively proven and Parfit does not rise to this challenge.
Because Parfit rejects the noumenal freedom thesis and therefore is driven to the randomness claim about free actions he reaches the conclusion that there is nothing to the claim that we could deserve to suffer. Instead Parfit claims all we can justifiably want is to make people come to understand the wrongness of their acts, not to aim at making them suffer. So Parfit rejects retributivism and adopts a kind of deterrence theory of punishment though, in doing so, he fails to address the relationship of this to an account of autonomy. Interestingly, however, Parfit seems to object to retributivism on the grounds of its affinity to consequentialism, something that would require a long argument with him if he had justified this claim in any detail which he does not.