It looks like the perfect workaround is to use some middleware to autogenerate the Swagger docs, at least to keep code and documentation in sync. After a while I found myself doing the opposite, using Swagger to autoconfigure endpoints (https://github.com/krakenjs/swaggerize-express) and even mongodb models (https://github.com/pblabs/swaggering-mongoose). Here is my actual reference architecture: https://github.com/pibi/swagger-rest-api-server

Best thing about this approach is the clean separation between API definitions and the implementation (all over the stack), so the teams can just discuss about how to organize the resources and how to use them.