Last Wednesday at 2 AM, a customer message woke me up. They asked if I could change a delivery time. I groggily replied three characters and went back to sleep. The next day I realized — this kind of thing happens four or five times a month.

That night, I started seriously looking for "something that stays awake for me"

OpenAI quietly launched a new product called Dots. In plain language: while you sleep, it stays "on duty" for you online. It can hang next to your inbox, customer service window, and calendar — only pinging you when something actually needs a human brain.

My friend Sister Lin (she does Xiaohongshu operations in Shanghai, running 8 accounts solo) told me she's already trying it. 80% of her repetitive DMs — "when does it ship," "how do I get a refund" — Dots handles automatically. She only looks at the 20% that truly needs a human.

Replication cost — I did the math myself

Money: OpenAI offers new users a free credit (check the official page for the exact amount). I used it for three days and didn't spend anything.

Time: From signup to having an agent start watching messages, I spent about 40 minutes. I did get stuck once — on the OAuth step (that's "authorize a third-party account to log in") I checked the wrong authorization scope. Starting over fixed it; this step cost me 20 minutes the first time.

Technical barrier: Zero lines of code, but you need to know how to authorize third-party accounts.

First step: Go to the OpenAI site, find the Dots entry, click Sign in, then choose Connect Gmail or Slack.

Honestly, if you're not drowning in messages every day, you probably don't need this tool. But if you're one person doing three jobs, it really helps you sleep peacefully at night.

Three types of people — my different advice for each

Just starting out (no real customers yet): If you're just beginning, don't bother with this yet. Right now every message you personally reply to is relationship-building — worth more than an AI reply.

Have 1-2 stable clients: If you already have steady clients, give it a try. Pick the most annoying repetitive question and let it reply, then check the data after a week.

Scaling up (team of 3+): If your team already has 3+ people, evaluate it seriously, but think clearly about this — customers have very different tolerance levels for "AI replies" vs "human replies." I'm still figuring this part out myself.

My judgment on it isn't complete. But that feeling of being woken up that night — I remember it too clearly. So I'm sharing it now.