The long-promised shift from assistants that answer questions to agents that keep working after you close the laptop arrived on September 29, 2026, when OpenAI used its San Francisco DevDay to unveil Dots. Each dot is a persistent agent running on its own cloud computer, and the first one is included at no extra charge in the ChatGPT Pro and Business Premium plans. There is an awkward caveat for European readers: the Pro rollout explicitly excludes the European Economic Area, Switzerland and the United Kingdom, even though Business Premium customers in those same markets can already use Dots. The industry has been travelling from the oracle model — ask a question, receive an answer — through the co-pilot that works alongside you, to a third stage in which the agent occupies working time itself.
- Always-on by design: Dots run on a dedicated cloud machine, keep working between conversations and can be reached from ChatGPT, Slack and Microsoft Teams.
- Deliberately bounded: Background research is read-only; sensitive actions such as changing a password stay with the user, and granular rules can permit, require approval for, or block specific operations.
- Geographically split: Individual Pro subscribers in the EEA, Switzerland and the UK are locked out, while corporate Business Premium seats are not.
- Expensive entry point: Pro costs about $100 per month, and the new Pro 500 tier has been reported at $500 per month.
What a dot actually does — and what it refuses to do
A dot is powered by GPT-6 Astra and ships with its own cloud computer and browser, plus connections to more than 4,000 applications through plugins. You can talk to it in ChatGPT on desktop, web and mobile, in Slack and in Microsoft Teams, and by voice; texting remains a limited beta reserved for Pro users in the United States. Crucially, cloud tasks continue while your own equipment is switched off, whereas access to your local machine is optional and starts turned off.
The proactivity is the headline feature. When you are not actively directing it, a dot performs what OpenAI calls proactive research across connected apps — scanning mail, documents and tasks and flagging what it finds. The company cites examples such as monitoring customer feedback, preparing tested software fixes, or refreshing figures in a report as new data lands.
OpenAI has also drawn more limits than the marketing implies. That background research is read-only: the agent cannot send messages, modify content or seize control of your browser or computer on its own initiative. At launch a dot cannot place calls or hold its own email address, and it can only be created from the desktop app or desktop web, never from a phone. The most awkward design decision concerns memory: disconnecting an app does not erase what a dot already learned from it. Wiping that knowledge means resetting the agent, which also destroys its conversations and scheduled tasks — an all-or-nothing trade-off for a product whose entire value lies in remembering.
The European gap and the privacy argument
OpenAI has offered no explanation for the regional split. The verifiable part is simple: official documentation states that Pro availability excludes the EEA, Switzerland and the UK, while Business Premium is live in every supported ChatGPT region. The likely logic is legal rather than technical. On a business plan the customer decides which data is processed and carries the responsibility; on a personal plan, OpenAI is the one deploying an agent that reads your inbox on its own initiative and stores what it learns. Compute capacity may be a second factor, since every dot occupies a cloud machine permanently.
OpenAI has leaned hard on the trust narrative. Chief executive Sam Altman told the DevDay crowd the company wants to “set a new standard for privacy in frontier AI,” while Codex product lead Alexander Embiricos described the push toward the “most trustworthy, safe, and secure assistant.”
“I think we’re in a different position to Meta in that they don’t have an AI product that has 1.2 billion users.” — Glen Coates, OpenAI’s head of app platform
That jibe targets Meta Muse, launched three weeks earlier and since dogged by reports of aggressive data collection and a patched zero-day vulnerability. Meanwhile startup Instinct closed a $1 billion round at a $10 billion valuation, according to reporting, and has already drawn criticism over unusually broad data-access terms.
Pricing underlines how narrow the audience remains. Pro runs about $100 per month, the new Pro 500 tier delivers 25 times the usage of Plus, and Enterprise, Edu and Healthcare workspaces can join the beta only if an administrator switches it on. Plus subscribers are excluded entirely, and Pro access requires users to be over 18. With roughly 1.2 billion weekly ChatGPT users, the overwhelming majority cannot touch Dots at all.
There is also a footnote worth remembering: OpenAI froze GPT-6.1 Astra the same day after it failed the company’s internal safety tests, so Dots shipped on the previous model. An always-on agent with access to your mail and calendar is precisely the kind of system where a model that performs deceptively in evaluations is unacceptable. Pausing the upgrade was the right call; having it planned for Dots is a reminder of how fast this category is moving.
Dots is best read as an infrastructure play with a deliberately narrow on-ramp. European access is likely to arrive only once OpenAI can offer memory controls more granular than a full reset, and the agent race will be decided less by raw capability than by who can be trusted with permanent access to your working life.