In the context of category theory, it is a general phenomenon that
On the other hand, a “nice object” is, loosely speaking, an object in some context which has more special properties than the generic object in that context will have. For instance manifolds are nice objects in the context of generalized smooth spaces. Fields are nice objects in the context of commutative rings.
Clearly, the more extra properties one imposes, the less likely it is that these are preserved under limits and colimits. For instance
not all quotients of a manifold by the action of a group are again manifolds (this is a colimit which fails to exist);
not all fiber products of surjections of manifolds are again manifolds (this is a limit that fails to exist).
it is often false that the coproduct of fields (in the category of commutative rings) is again a field.
Without making any sweeping judgments, a general nPOV heuristic is that faced with a choice between working with a non-nice category of nice objects and a larger nice category of non-nice objects, it is usually preferable to switch attention to the nice category. For example, instead of working with the category of smooth manifolds, work instead with an ambient category with better properties, such as a category of diffeological spaces or a model of synthetic differential geometry, in which the category of manifolds is fully embedded. Moerdijk-Reyes can be read largely as an implementation of that philosophy.
On the other hand, another methodological heuristic is to move from a nice object to a nice category attached to it. For instance, the category of modules over a field is about as well behaved as it could possibly be, and similarly one often contemplates the category of vector bundles over a smooth compact manifold.
The notion of “nice objects” can be formalized to some degree for instance in terms of Isbell self-duality as described in Lawvere.