There's definitely a distinction between fetching data for training and fetching data as an agent on behalf of a user. I guess you could demand that any program that identifies itself as a user agent should be blocked, but it seems counterproductive.

Counterproductive for what?

If I am writing for entertainment value, I see no problem with blocking all AI agents - the goal of text is to be read by humans after all.

For technical texts, one might want to block AI agents as well - they often omit critical parts and hallucinate. If you want your "DON'T DO THIS" sections to be read, better block them.