The problem with the neo-Humean “just so” account of causation is that we can’t speak of things acting this way or that without being able to specify which things we mean; and the point of final causality is that things are defined in terms of their proper powers and effects. Meaning, the neo-Humean lacks an intelligible criterion for saying that some things T’ just happen to act in a T’-like way, while other things T* just happen to act in a T*-like way, since it is only in virtue of the fact that those disparate things naturally act for their proper ends that we can separate them in the first place. Grue and all that. In order to exist as a T’, rather than as a T*, the thing would have to include T’-like finality, otherwise it could just be a T’* going through its T’-like phase.
That’s how grue shows Humean causality is incoherent: a merely stochastic or empirically familiar account of causality offers zero insight into a real thing’s proper place in the larger casual nexus. You cannot coherently say that O is a red object while also claiming that O merely happens to reflect photons at a wavelength of 630–750 nm; to possess that causal power in the visible spectrum just is what it means for O to be a red object. Nor can you say that some A is an apple seed, but merely happens to result in an apple tree, for, again, A‘s existence qua A is inseparable from A’s final end qua apple-tree-forming. If you said that an apple seed could just as easily turn into a hippopotamus, you would not be talking about a real apple seed. So, we should either count a thing’s end to be as important for knowledge of the thing as its essential attributes are, or give up on causation entirely. In a way, the latter is what Lewis did by jumping the shark with his charmingly absurd actualism.