Dots

OpenAI introduced Dots at its DevDay event on 29 September 2026. Every Dot is powered by GPT-6 Astra and given its own cloud machine, with a browser, a terminal and storage for files, so the work continues once ChatGPT is closed. Entry costs $100 a month through ChatGPT Pro; companies can instead buy Business Premium, likewise $100 a month per seat when paid yearly. Individual Pro subscribers can use it in every ChatGPT market bar Britain, Switzerland and the countries of the EEA.

Morning, with a colleague in the cloud

Setting one up happens inside ChatGPT, in the browser or the desktop app: you pick a name, settle on a cartoon look and then treat the result much as you would a new colleague, typing to it or ringing it for a voice call. The same thread of memory follows it into Slack and Microsoft Teams. While you get on with other things, it quietly reads through the apps you have linked, never with permission to change them, and comes back with proposals. In one early case it noticed a freelance piece nobody had billed, prepared the invoice from the relevant emails and held it until the user said yes.

Where it earns its keep

Reach is the headline. Through OpenAI's plugin catalogue a Dot can work with upwards of 4,000 apps, from Gmail and Google Docs to GitHub, Slack and QuickBooks, and it passes programming jobs to Codex. Initiative is the other selling point: rather than waiting for instructions it proposes work, yet its background exploration cannot send or alter anything. Custom Rules let you permit an action, demand approval for it or forbid it outright. Saved passwords stay hidden from the model even while the Dot uses them, and conversations with your Dot are not counted against ChatGPT usage limits.

The catches

Price is the first obstacle: nothing cheaper than the $100 Pro tier comes with a Dot. Europeans paying personally are shut out for the moment, since Pro subscriptions sold in Britain, Switzerland and EEA countries arrive without one. Chatting costs nothing against your limits, but any job the Dot launches in Codex or ChatGPT Work is drawn from your ordinary allowance, and OpenAI has given no figure for the bill that heavy proactive use may run up. Timing has not helped either: Dots landed in the very week OpenAI disclosed a string of incidents with its research agents.

Keeping the leash short

Protection comes in layers. OpenAI's own policies sit beneath your Custom Rules; an automatic check examines any step that would touch an account or pass information on; a monitor can pause the Dot; and the Activity View keeps a ledger of what it has done. Password changes are always left to you. Workspaces on Business, Enterprise and Edu stay out of model training by default, personal accounts follow whichever ChatGPT training choice you made earlier, and OpenAI says the private notes and background research a Dot produces are not fed straight into training. The real exposure is breadth, so link only what a task requires.

For a ChatGPT Pro subscriber outside Europe, Dots is the most capable operator in the field and the one most willing to find work unprompted. Write your Custom Rules before connecting an inbox, and extend write access only as the agent proves itself.