Work · the AI in the inbox
AI auto-replies
30-second gist~30s read
Outlook, Gmail, and Apple Mail can now read an incoming email and propose a polished reply in your tone. It's very fast. It's also lossy: the AI optimises for "sounding like a reply", not for "saying what you actually meant".
Use it for routine logistics. Don't use it for relationships, decisions, or anything political at work.
If you want more
The classic auto-reply mistakes
- Confidently agreeing to something you wouldn't have agreed to. The AI can't read between lines about commitments.
- Polished response, wrong audience. Reply meant for one person sounds odd to a wider group.
- Politely declining something the relationship needed you to say yes to — the AI doesn't know what's social.
- Adding a closing flourish that wasn't your style ("Excited to chat further!" if you'd never write that).
A two-second habit
Before sending, read the AI draft as if it had been sent to you. Does it feel like you said this? Does it commit to anything? Has it missed the actual ask? Most of the bad outcomes here are caught by a single ten-second re-read. Most of the bad outcomes happen because the AI looked done, and we trusted that.