To me, the problem is that “Serverless” is trying to shoehorn HTTP into an RPC protocol instead of harnessing some of the decisions, like resources, that went into the design of the HTTP protocol.
I think it can still fit together if the serverless functions act as the middleware that does the processing. Another could then act as the actual “service”. The important part for me was to look at APIs in the context of resources, a uniform interface and workflows (middleware) to control the flow. How it all fits together depends on the technology of choice. Maybe this is even an open protocol vs. proprietary platform question. Serverless isn’t really a programming language or protocol but a proprietary service.