OpenAI's new dots are agents with their own computer. Ask three things first.
30 Sep 2026 · News
OpenAI's big new thing from DevDay is a dot: an agent with its own computer that keeps working after you close the chat. Three questions before you hand one your week.
At DevDay on 29 September, OpenAI announced more than 20 things. Plenty of it is for people who build software. The piece most readers will meet first is dots.
A dot is an agent that keeps working when you are not in the room. OpenAI calls dots "always-on agents". Each one runs on GPT-6 Astra and gets its own cloud computer, its own browser, and access to over 4,000 apps through plugins. You can open that computer to see what it did. You talk to it in ChatGPT, Slack, or Teams. OpenAI's own example: an early tester's dot noticed he had forgotten to invoice a publication, prepared the invoice, and sent it after his approval.
If that shape sounds familiar, it is close kin to Meta's Muse. Two labs, one idea: give the agent a computer of its own and keep a human near the send button.
I have not used a dot. What follows comes from OpenAI's own pages. Here are the three questions I would ask first.
1. What can it do without asking?
Dots start with built-in rules. Per OpenAI's safety FAQ, changing a password or moving money always goes back to you. Permanently deleting data or installing software may need your approval each time. Recurring messages can be approved in advance. Custom Rules let you allow, require approval for, or block specific actions. A separate check called Auto-review looks at planned steps like sending email before they run.
The practical tip is in that same FAQ: be specific when you pre-approve. Who gets the message, what it says, and when it goes out.
2. What does it remember, and can you see it?
A dot shares memory with ChatGPT and can keep context for as long as you keep the dot. Today you can delete a dot's context by deleting the dot, but you cannot view or edit individual memories. In the background, a dot also does "proactive research" on the apps you connect, using read-only access. So connect fewer apps at first, and add more once you trust it.
3. What does it cost after the first month?
The first dot is included in Pro and Business Premium at no extra cost, with extended limits for the first month after launch. Chatting with your dot does not count toward your usage limits. Tasks it hands to Codex or ChatGPT Work do. OpenAI says you will be able to add more dots, or raise each dot's speed or monthly workload, in the future. It has not said what that will cost.
Also check whether you are eligible: Pro outside the EEA, Switzerland, and the UK, plus Business Premium. Enterprise gets a beta that admins switch on, off by default. Rollout is gradual, so it may take days to show up.
The rest of DevDay was a long list of ChatGPT, Codex, and developer updates. One pricing note from the same day: GPT-6.1 Sol is a model priced at one-fifth of Astra's standard token rates.
OpenAI's own caveat is worth repeating: dots "can still make mistakes", so review anything consequential.
Simple rule: Before you give an always-on agent a job, find out what it can do without asking, what it will remember, and what it costs when the launch allowance ends.