REP27Eleven at night. A client in another country writes to ask whether the law applies to him. Nobody is awake. The message sits there until morning, and by morning he has written to someone else.
I built the thing that answers instead. Then I spent longer deciding what it is not allowed to do.
Three settings. One of them is the whole point.
It can draft and wait for me. It can send on its own only when it is confident, and hold everything else as a draft. Or it can send directly. I choose per mailbox, and I can turn the whole thing off with one switch.
Some words stop it dead. Refund. Chargeback. Lawyer. Complaint. If one of those appears, the reply is never sent on its own — it waits for a person, and that person is me.
It does not answer instantly either, because nobody writes a considered legal answer in four seconds and clients notice. It waits half an hour, adds a few random minutes, and only writes inside working hours.
It replies in the language the client wrote in, out of eight. It ends with one useful question about that specific case, never a generic "any other questions?". And when someone says they want to proceed, it builds the real link — the prices are the ones on the website, never invented, and a second check runs before anything leaves.
Every reply it sends lands in my inbox as a hidden copy. I read them. That is not a formality: it is how I know whether to widen what it is allowed to do or narrow it.
So here is my honest question, and I would like a real answer, not a thumbs up. Would you let an AI answer your clients? And if yes, where exactly would you draw the line — price questions, legal questions, complaints?
Tell me in the comments. I read every one and I will tell you what I decided and why.
