interesting / I doubt the agent will read 70 pages, because the AI models would hit token limits. Do you know the word count?
In case it's hitting the token limits >
What you should do is in your doc, if you've different overall themes for eg: first half talks about marketing, 2nd half talks about sales, divide the content that way.
Then run an AI agent that would first scan and understand the context of the lead magnet and only pick snippets from the overall theme.
So the flow would be like this: first, read the knowledge base, then an AI agent step will check for the context. Then there's a condition that will say something like, if the context is sales, another AI agent will have a prompt to create lead magnets. The other condition would be, if the overall theme is sales, then another AI agent step will have a prompt to create a lead magnet. That is how I think it will work if we are hitting the token limits.