I agree, I especially find it useful when prototyping. I recently built a canvas node-based tool that connects LLM prompts and passes the model's response into the next node as context. It was fun to create and great for testing/comparing outputs from different models. https://littleworkflow.com/