Inbox & Communication

Inbox & Communication

The Quick-Thanks Trap

AI is unnervingly good at writing replies you shouldn't be sending at all.

A client emails: "Thanks, got it. Talk next week." Nothing is owed; the thread is closed. But you ask Cowork to draft a warm acknowledgment anyway, because the draft is right there and free. Two seconds later you have three gracious sentences. You hit send.

Then you do it fourteen more times that morning. Inbox to zero. You feel fast.

By afternoon the replies to your replies arrive ("No problem!", "You too!") and you're acknowledging acknowledgments.

Cowork didn't make a mistake. It wrote exactly what you asked for, and wrote it well. That's the whole problem. A good draft makes a message you shouldn't send feel like one you already decided to send.

The fix isn't a sharper prompt. It's a gate before the prompt.

#The bottleneck was never the writing

Think about what used to stop you from firing off a reflexive "thanks." Not the effort of typing eight words: that was always trivial. What stopped you, on a good day, was a half-second of judgment: does this thread need another message, or am I just twitching?