I think the "free will" level is to the "neural activity" level as thermodynamics is to detailed models of physical processes. In thermodynamics we sort of forget about the detailed mechanics going on "under the hood" and talk in terms of quantities like "temperatures" and "entropies". If we look sufficiently closely these quantities dissolve away and we end up talking about the constituents of the system in more exact physical ways. A lot of the time, though, the thermodynamic approach lets us summarise situations and model their behaviour in much more succint ways. Similarly, all this talk of intentional agents and free will and choices is a free-floating explanatory level that lets us talk concisely about people's internal states and behaviour. The "real" states and dynamics are all down at the level of neural activity, and if we look ever closer to that level (or beyond it to the dynamics of molecules) then the intentional agents with free will dissolve into the mist. People who forget this and talk about free will as an irreducible force in the universe are, I think, making a fundamental mistake - they're confusing useful shortcuts made by our crude, coarse-grained models of people with reality.

Furthermore, I think that talking in terms of a free-floating intentional level lets us more easily consider such matters as free will in systems with other low-level substrates - fictional characters, animals, aliens, AIs... Without having such a clear view of things I'd imagine we'll get into all kinds of muddles.