On 29 September OpenAI introduced dots: always-on agents powered by GPT-6 Astra that have their own cloud computer, remember context and work in the apps you connect. They start with Business Premium and with Pro, but Pro is for now outside the European Economic Area, Switzerland and the UK.
- You talk to a dot in ChatGPT and through Slack. With permission it uses your computer too, and that access starts switched off.
- When it is not on a task, it does "proactive research" in connected apps. The tools it uses for that cannot directly send messages to other people, change content or control a browser.
- Under the rules a dot is given, changing a password and transferring money require the person to finish them.
You can ask it to review your calendar every morning and tell you what is coming up that day.
That is the example OpenAI gives itself in its help pages, and it is well chosen. Not code, not research. Dull work everyone has let slip at least once.
Now the other half. An agent with its own computer, access to your mail and to your apps, working while you sleep, is a wonderful employee and a perfect door. And a large part of the help pages is about control: rules for when it acts alone and when it asks, review of its actions, an automatic check before some of them.
Here is what gets skipped: proactive research. When you have not given it a task, the agent reads through your connected apps on its own, looking for ways to help. It cannot send or change anything from there, OpenAI writes. Sounds harmless, until you remember how much there is in one inbox. And what it learned does not go away when you disconnect the app, only when you delete the dot itself.
If you try it, start with few permissions and watch Activity View for the first week. Not because a dot is bad. Because you do not know it yet.