The familiar AI chat pattern is simple: ask a question, receive a response, close the window. OpenAI’s Dots announcement at its 29 September DevDay pushes in another direction. The company presented Dots as agents that can stay active and work on ongoing tasks for a user, according to Associated Press reporting from the event.
The distinction matters. A continuing agent may need to remember a goal, decide when to act and interact with connected tools. That can make repeated chores easier, but it also makes permissions and review points more important than they are in a one-off conversation.
What was announced
AP reported that chief executive Sam Altman introduced Dots during the company’s annual developer conference. OpenAI’s own updated GPT-6 Astra system card says Dots are built on existing model and agent capabilities and powered by GPT-6 Astra. The company describes them as always-on agents. That is a statement about the product’s intended operating pattern; it does not establish that an agent can complete every task reliably or should be trusted with every account.
For a reader, the practical questions come before the science-fiction comparisons. Which apps may an agent access? Can it spend money, send messages or change files? What must it show a person before acting? Those are the controls that determine whether an always-active helper feels useful or intrusive.
A busy week for agent safety
The launch follows a separate decision by OpenAI to hold back another advanced model over safety concerns, AP reported. The two developments underline that capability and safe deployment are related but separate decisions. A feature announcement does not remove the need to test it in ordinary use.
For now, Dots is best understood as a product direction unveiled at DevDay. Readers should look for clear availability details, permission settings and independent hands-on testing before deciding whether it belongs in their daily workflow.
Sources & further reading
Prepared with AI assistance from linked reporting. The cover is an AI-generated editorial illustration. Spotted something we should correct?