The definition of an agent is so vague now. What is the agentic promise and why wouldn't a workflow tool with LLM steps cannot deliver on this promise? Also what vulnerability are you referring to? I have no stakes in n8n or node red, but being able to embed LLMs in a workflow and conversing with a "workflow" seems to solve a lot of agentic/assistance problems (Make a report of revenue, and send an email report to these people,for example)

What is the completely different approach you are suggesting?