Then you can be writing the code, with the LLM doing the "boring" parts, in chunks small enough you can review them on the fly.
