Which is basic: our evaluation of the world as a whole, or our evaluation of its individual constituents? Are we to work up from my particular desires to find which total state of the world I should prefer? Or do we instead start with my global preferences (a value function over possible worlds) and abstract away various details to yield my more particular desires?
An example of the latter approach might be Liz Harman's analysis: S desires that P ("all things considered") iff S prefers the nearest P-world to the nearest not P-world. One worry: on this view, it no longer seems that we can assign quantitative "strengths" to desires. (And aren't numbers elegant?)
But perhaps the upshot of the holistic picture is just that individual desires aren't really so important anyway. Utility values may be assigned to whole worlds, and combined with credences to yield expected utilities, etc. So it's still an elegant picture, and one which fits well with formal decision theory. It's just that the value of a whole world-state is not reducible to any simpler description of the values of its parts. (For example, you can't just sum all the pleasure and pain if it turns out that sadistic pleasure detracts from, rather than adds to, the overall value of the state of affairs.)