I agree, we should separate out when the LLM is used for orchestrating and when it is used for processing information. The more the two are tied together, the harder is it to debug.

→ View original post on X — @jiquanngiam,