As I said in another post, I can't find the workflow for this one. I thought if I found it, I would answer my own questions. But, as I have continued to follow the YT video, I don't think having the workflow would help that much. So, here are my questions. 1. What is the Loop over items/Google download doing? The loop was added in this version. But wouldn't the workflow Set File Id -> Download File -> Switch work without a loop? 2. In the Switch node, how did you know to set, for example, application/vnd.google-apps.document? Where did you find the vnd.google-apps... part? 3. Where does the data for the Enhanced Default Data Loader come from? This part {{ $json.data || $json.text || json.concatenated_data }} I love how the workflow has evolved and now purposefully works with different file types. The improved metadata is helpful for real applications.

Anyone having success with using a SerpAPI as a tool for an AI Agent? When I use it, either the AI Agent doesn't use it for the question in chat, the AI Agent ignores the result from SerpAPI, or the result is bad or not useable. I have use cases where I want to use Google search results as part of a longer workflow. But so far, I can't get anything useable from SerpAPI. A related question: Can you "force" an AI Agent node to use the attached tools? (Perhaps in the system prompt.)

The AI Agent prompting video was very helpful. Thanks for putting it out there. I'm coming from Make so I want to ensure I understand where to put the prompt. In the Make.com openAI modules, you can add system, user, and assistant messages. (Make.com doesn't have the equivalent of an AI Agent node.) From what I see, the n8n AI Agent node only has a system message. So, do you put the entire prompt (e.g. a long structured framework prompt) into the system message?