Dots is the headline from DevDay, and the interesting part is not the persistence but the hardware. Each dot runs on GPT-6 Astra and is given its own cloud computer and browser, kept separate from yours unless you connect them. It holds several projects at once, reaches more than 4,000 apps through plugins, sends progress updates and asks questions as it goes. You reach it in ChatGPT on desktop, web and mobile, or message it in Slack and Teams.
You can watch the screen while it works. That sounds cosmetic and is not: the standard complaint about handing a task to an agent is that you get a verdict with no way to tell where it went wrong. A visible screen turns an opaque result into something you can supervise.
Rollout is uneven in a way worth knowing before you plan around it. Pro and Business Premium in eligible markets, with Pro excluded in the EEA, Switzerland and the UK at launch; Enterprise, Edu and Healthcare get a beta once an admin enables it. OpenAI's own framing is that dots can still make mistakes and consequential work should be reviewed — which, for an agent that runs while you are not watching, is the whole design problem in one sentence.
Read it alongside The agents got their own computers: three vendors made the same architectural move this quarter.