One of the major gripes for me about UML diagrams as I've seen them used so far is that they mostly only tend to serve as "pretty pictures" on someone's wall or as a folded page in a binding somewhere and rarely serve as a baseline for a production code.

In this type of scenario - the UML diagrams (or whatever flavor of modelling language you use) are rarely useful in the long run, as they tend to suffer from loss of relevance as the time goes on and the project evolves. These initial drawings are mostly useless and incorrect in few year's time and having them around is mostly harmful.

That said, using modelling tools (not necessarily UML) is not entirely useless practice, if the models provide actual live and relevant input to the actual running code. If the model description is a useful artifact of the code, it should be treated as code:

Either by using it as a direct source of metadata for making dynamic decisions based on the modeled concepts, or using code generation to generate static code artifacts that get used in the actual working code.