so first off, i'm not an AI avoider - i use LLMs all day at work, and have several personal projects i'm likewise using claude for. but a single, interactive session produces code as fast as i can handle, if i want to keep a tight rein on the quality of the output - i often find myself having to look at the code generated from some intermediate stage and course-correct.

i do use agents for code review but again in a very human-in-the-loop manner; i will ask for a review and then read it over before feeding it to my coding LLM session and asking it to read and iterate on it. as you say, it can be surprisingly tricky to get right, and i like the confidence of the human being the coordinating reviewer.

thanks for all the examples, though - that does give me a lot more context for what people mean by "software factories" (and the assurance that they don't fit the way i work).