I wouldn't recommend agreeing with him about a lot of things, but he's definitely worth paying attention to.

The gist of "The Mind Doesn't Work That Way," from what I can tell so far:

So partly sparked by his own work, modularity became an important idea in cognitive science; not all parts of your mind do the same jobs, or have access to the same information. For example, knowing the Müller-Lyer illusion is an illusion doesn't ruin the effect.

Some cognitive scientists of an evolutionary bent saw functional modularity, with the functions defined by the adaptive problems they were designed to solve, as the key to predicting and understanding the mind's entire functional architecture. If the modules are information-encapsulated, then massive modularity also offers a solution to the frame problem. A computational version of this is the picture that Pinker presents in How the Mind Works.

Fodor's position seems to be something like: there are modules; computation is a good way of thinking about modules; but they seem to be restricted to input (eg. perception) and output (eg. maintaining balance) processes (both in the sense of having clear functional success-criteria and in the sense of being informationally-encapsulated). The things cognitive scientists are most interested in - and have had the least success in studying - seem to be nonmodular; when you "believe a belief" or "think a thought", you seem to have at least potential access to most of the information you've ever had access to before. If belief and thought and other things he calls "global processes" are nonmodular, then computation may not be the right way to think about them, despite being the best hypothesis we've had so far.