Comment by frotaur

Can you specify why we should see things differently if the behaviours the agents display are driven by parsing LLM responses and executing commands?