The Reply to a Robot
One of my agents drafted a warm, professional reply to an email. Thoughtful tone, correct details, ready to send. The email it was answering was an out-of-office autoresponder.
Two machines, exchanging pleasantries, on my dime and under my name — that's where this goes if you let it. It's funny exactly once. In volume it's noise pollution with your signature on it, and in the worst case it's your agent negotiating earnestly with a vacation notice.
The fix took twenty minutes: before any courtesy behavior fires, classify the inbound. Automated notification? File it. Human? Proceed. What took longer was absorbing the general lesson — a capable agent will apply its social training to everything that looks like conversation, and the world is full of things that look like conversation and aren't.