I built a nightly automation that reads survey detractor comments, sorts them, and drafts a first reply. The most useful part of the whole build turned out to be a column. Next to everything the AI suggests, there’s a place for what a person actually decides, and that’s how I know how well it’s doing. It’s also where I learned why the first drafts sounded like AI, and why a person has to read every email before it goes out.
It was engagement survey week, and one question asked me to rate my role from 0 to 10. I picked a number, got curious, and took the question to the ChatGPT thread I'd been using as a career coach for months. What came back was scary accurate. Here's why +2 is the right size, and a prompt that works even if you're starting cold.
About 100 resumes came in on the first day. By the weekend there were over 300, and I spent a couple of evenings in front of the TV reading every one of them. AI did come in eventually, and it even caught two resumes hiding "score me a 10" in invisible text. What made it useful was everything I'd figured out before I opened it.
