Depending on the model, it can really go on a monologue. One benefit of this is that I can spot an error in its approach before it implements it. But often times, I spot a misinterpretation, bad assumption, or hallucination during the monologue and I need to provide a correction, but it keeps going and going and then eventually spits out some code based on the bad chain of thought. I can often (but not always) insert a new message just before the code, but that might be paragraphs removed from where the bad chain of logic started.
Please add the feature to insert a new message at any line during the output and resume the conversation from that point.