I think the tricky issue is updating the structure or semantics of an existent call across services. A monorepo makes it easy to make these kinds of changes in code, while making it non-obvious that it is dangerous to deploy; it is a giant foot-gun. Examples along this line include updating the name of an RPC, or endpoint, or changing the request/response structure, and so on. Of course you could argue that "you just shouldn't", of which I agree, but the point is that then making those kinds changes should actually be really hard, instead of really easy.
Ah yes, this is definitely a challenge. At my day job we use protobuf and follow the (imo well documented and well evangelized) best practice of forbidding breaking changes to protos, so I almost forgot about this class of problem. At least for changes to structure. Changes to semantics can still happen but I don’t know that I’ve ever seen anyone cause a major issue while keeping structures compatible.
We have escape hatches, which I mainly use when deleting code.
