[question] What magnitude of training data do we need to use AgentFlow?

[vanetreg]

I processed your repo, docs and Deepwiki pages in NotebookLM and I was told we need 10k+ data points to train the system properly.
Is this true?
This can be a huge barrier for many of us,
so pls. clarify, when and under what use cases and circumstances it is / it is not true!
Thank you! :)