This doesn't necessarily sound bad to me. The natural language being passed around is effectively an information transfer protocol. Each agent has instructions for how to manipulate and respond to messages, and the publicly-posted final document is akin to HTML source, in that a user's agent interprets it for consumption rather than display it directly.
The only problem is lossyness and hallucinations, but assuming the technology improves such that that becomes a non-issue then having bots deal with formalities and formatting sounds great.
One could also flat out refuse to talk to bots. It might seem childish but in the long run could be fine saving considering how people will use them to water your time and confuse you.
It's like people skipping preselection tired topic menus in self service via phone.
The only problem is lossyness and hallucinations, but assuming the technology improves such that that becomes a non-issue then having bots deal with formalities and formatting sounds great.