Would you let AI send your emails without seeing them first?

Genuine question, not rhetorical.

Every AI writing tool is drifting toward autonomy ... read the thread, draft the reply, send it, tell you afterwards. The demos are genuinely impressive.

But the failure modes aren't symmetrical. A bad draft costs you ten seconds. A bad sent message costs you a client, or an apology, or a friendship. Or ... your reputation.

Every time I've asked, the answers split cleanly in two:

1. "Absolutely, for the boring stuff. Confirmations, scheduling, thanks-got-it."

2. "Nothing goes out under my name that I haven't read."

Almost nobody sits in the middle. And the people who say yes are usually picturing one narrow category of low-stakes message, not their actual inbox.

So where's your line? And does it move if the thing has been right two hundred times in a row?