I get the author's points, and they all are valid, but I don't understand why people would use generated code and its types through their entire project.

This is just asking for trouble when the API will inevitably break as all APIs will do eventually. In our projects I mandated and pushed really hard that we create intermediary data classes that correspond one to one to the protobufs (at first).

I got a lot of angry faces and reactions in PR due to the seemingly useless boiler plate code required but it saved our butts so many times when the API changed just before a release that it became the de facto standard.

Also, protobufs and GRPCs are a de facto standards. Are there better alternatives? Yes. Should you use those? Most likely not because the point of serialization frameworks is to be used by many people in various tech stacks.