At DevDay on September 29, 2026, OpenAI introduced dots: agents that do not wait for a question to answer, but work autonomously on goals the user assigns them. Each has its own computer and browser in the cloud, connects to the user's apps and writes in ChatGPT, Slack or Teams only when a decision is needed. It is OpenAI's clearest move yet toward an assistant that acts instead of just chatting.
How they work
According to OpenAI the mechanism is simple: the user gives the dot a goal, connects the apps it needs and sets what it can do on its own. The dot works and brings results back for review, learning from the user's feedback over time. The examples cited range from booking a table to far more ambitious projects.
The main features:
- Its own computer: each dot works from a cloud environment with its own browser
- Over 4,000 connectable apps, through ChatGPT plugins
- Messages where you already work: ChatGPT, Slack or Teams, when an approval is needed
- User control: you decide which apps to connect and which actions the dot can take without asking
According to OpenAI, each dot uses GPT-6 Astra, the flagship model introduced on September 3.
Who can use them and what they cost
Dots are rolling out to Pro and Business Premium subscribers in eligible markets, with a difference that matters directly to readers in Europe:
- Pro plan: the rollout excludes the European Economic Area, Switzerland and the United Kingdom
- Business Premium: dots arrive in all supported ChatGPT regions
In practice, for now a Pro subscriber in Europe does not have them, while someone on a Business Premium plan does. The first dot is included in the plan at no extra cost, with extended usage limits for the first month after launch.
The demo that stumbled
During the live presentation, Holly Li from OpenAI's product team asked the agent, nicknamed "Dottie", for an update on the previous night's user tests. The dot acknowledged the request but stayed silent for about ten seconds; asked again, it replied that it was still checking. Li closed the moment with a joke: Dottie was having a "slow morning".
A small incident, but a telling one: reliability remains one of the central challenges of autonomous agents.
Why it matters
With dots the question is no longer how good an answer is, but how safe it is to let software act on our behalf, for hours, with access to apps and accounts. The more tools and autonomy an agent has, the more it matters to decide what it can do alone: a theme that, as we explained when covering Lockdown Mode and prompt injection, also concerns protecting data when an AI reads external content.
For a business, the practical question is to start with repetitive, low-risk tasks, such as summaries, reminders and research, and keep human approval on anything that touches money, customers or sensitive data.
Read also: Codex Cloud, OpenAI's reusable development environments and OpenAI launches GPT-6.1 Sol.