At present there is no consensus theory explaining the evolutionary stability of mutualistic interactions. However, the question is whether there are general 'rules', or whether each particular mutualism needs a unique explanation. Here, I address the ultimate evolutionary stability of the 'agricultural' mutualism between fungus-growing termites and Termitomyces fungi, and provide a proximate mechanism for how stability is achieved. The key to the proposed mechanism is the within-nest propagation mode of fungal symbionts by termites. The termites suppress horizontal fungal transmission by consuming modified unripe mushrooms (nodules) for food. However, these nodules provide asexual gut-resistant spores that form the inoculum of new substrate. This within-nest propagation has two important consequences: (i) the mutualistic fungi undergo severe, recurrent bottlenecks, so that the fungus is likely to be in monoculture and (ii) the termites 'artificially' select for high nodule production, because their fungal food source also provides the inoculum for the next harvest. I also provide a brief comparison of the termite-fungus mutualism with the analogous agricultural mutualism between attine ants and fungi. This comparison shows that--although common factors for the ultimate evolutionary stability of mutualisms can be identified--the proximate mechanisms can be fundamentally different between different mutualisms.