Currently, the LLM is not executing tools that are present in a parallel tool calling turn along with a talk_to_user call. We should do the opposite, rather drop the talk_to_user call to force the LLM running all the tool calls before returning to the user
Currently, the LLM is not executing tools that are present in a parallel tool calling turn along with a talk_to_user call. We should do the opposite, rather drop the talk_to_user call to force the LLM running all the tool calls before returning to the user