When working with monad transformers, I tend to stick to constraining functionality over unspecified monad. This essentially acts as a set (of constraints), which results in all the set-based woes that arise when effects are used. So not monad transformers themselves per se, you're right.