OpenAI Dots are beginning to roll out as a new kind of ChatGPT agent, designed to keep making progress after a conversation ends rather than waiting for the next prompt. A Dot has its own cloud computer, can work across connected apps and can run scheduled or proactive work while the user is elsewhere.
That changes the basic relationship people have had with chatbots. Instead of opening a chat, asking for a result and closing it again, a user can hand over an ongoing responsibility and return later to see what has happened. OpenAI says Dots are powered by GPT-6 Astra and are being rolled out gradually, so access and features will vary by plan and region.
OpenAI Dots are meant to own ongoing work
The important word in OpenAI’s description is not assistant. It is responsibility. A Dot can be given a goal, work out what needs to happen next and bring results back for review. OpenAI gives examples such as checking a calendar, researching a topic and running recurring tasks.
Each Dot has a cloud computer separate from the user’s own machine. Local computer access is optional and starts turned off. If a user deliberately connects a computer through the desktop app, the Dot can then work with files or a local browser, but that is a separate permission rather than the default.
The agent can also use connected apps. OpenAI says it may review information from those apps proactively and form memories from it, even when the user has not asked a fresh question. Slack and personal email can be connected, and scheduled work appears in the Dot conversation when it runs.
That puts Dots in the same broad shift towards agents that LiveAIWire has been tracking through products such as Microsoft’s more autonomous Copilot work. The common idea is that the AI does not merely answer. It watches, remembers, acts and returns when something needs attention.
The convenience comes with a larger trust decision
An agent that works between chats also receives more opportunity to make a mistake between chats. OpenAI’s own help page says users can create custom rules covering actions such as sharing, purchasing or access, with options ranging from acting without asking to handing the decision back to the user.
OpenAI also warns that a Dot can make mistakes even when following those rules. That caveat matters more for an always-on agent than for a conventional chatbot because the agent can be given connected information, recurring tasks and a longer window in which to act.
The company’s GPT-6 Astra system card added a Dots appendix on 29 September. OpenAI describes Dots there as always-on agents built on existing model and agent capabilities. The safety material also reflects the broader problem of giving capable models enough freedom to be useful without making oversight ceremonial.
This is why permissions and auditability are central rather than secondary product features. A calendar summary has a different risk profile from sending messages, purchasing something or changing a file. The user needs to know not just what the agent can see, but which actions it can take without asking again.
LiveAIWire has previously examined how agent access can expose private owner details. Dots push that privacy question into a more routine setting because useful ongoing assistance depends on context, connected services and memory.
Availability is deliberately uneven at launch
OpenAI says Dots are initially rolling out to Pro users outside the European Economic Area, Switzerland and the UK. Business Premium users can access them across supported ChatGPT regions, while Enterprise users can try a beta when a workspace administrator enables it. The rollout is gradual and OpenAI says it may take several days to reach an eligible account.
That means Dots are not yet a universal ChatGPT feature, and their launch should not be confused with a claim that every user can create one now. It also gives OpenAI time to learn from a smaller deployment before a wider release.
There are practical limits too. OpenAI says a Dot cannot initiate a phone call at launch. Texting is a limited beta where available. Creating a Dot currently requires desktop web or the desktop app, although users may later talk to an existing Dot through mobile access.
The real test is whether delegation becomes less work than doing it yourself
The promise of an always-on agent sounds obvious: give away repetitive monitoring and keep your attention for judgement. The harder test is whether users spend less time checking the agent, correcting it and managing permissions than they would have spent completing the task directly.
That is where Dots could either become a new layer of everyday computing or remain an advanced feature for people with unusually complex workflows. The product has the ingredients for persistence: a computer, apps, memory, schedules and a strong model. What matters now is whether those ingredients produce reliable follow-through.
The significance is therefore bigger than one ChatGPT feature. Chatbots trained people to think of AI as something they ask. Dots are an attempt to make AI something they assign. If that behaviour becomes normal, the next competitive question will be less about which model gives the best single answer and more about which agent can be trusted with a responsibility that is still there tomorrow.
About the Author
Stuart Kerr is Technology Correspondent at LiveAIWire, covering artificial intelligence, cybersecurity and the social impact of emerging technology. LiveAIWire is an independent, human-led technology publication using AI-assisted research, editorial production and original AI-assisted editorial illustrations under his direction.