Answer in brief
Dots are designed to continue tasks between chats. OpenAI's own setup guide also makes their permissions, regional rollout and need for user review part of the launch story.
An agent meant to keep a task alive
At its 29 September developer event, OpenAI introduced Dots, agents intended to make progress on a continuing goal after the user leaves the conversation. CBS News reported the launch as a move toward proactive, ongoing work. The product claim is broader than a reminder: OpenAI's help guide says a dot can work through a problem, choose next steps and return work for review. Whether that helps in practice depends on the boundaries of the specific task and the quality of the connected information.
What the setup guide actually permits
The OpenAI guide says a dot has its own cloud computer and can connect to apps selected by the user. It may inspect connected information proactively and retain context from it, so permission choices are consequential even between explicit requests. Local computer access begins disabled and requires a separate connection and approval in the desktop app. The guide also describes activity views for work in progress, scheduled and completed tasks, plus ways to pause or reset the dot. Those controls deserve a real review before an agent receives a calendar, mail or business system.
A launch with uneven access and unresolved performance
OpenAI says the rollout starts with eligible Pro users outside the European Economic Area, Switzerland and the UK; Business Premium and enterprise paths have different conditions. Gradual delivery means an eligible account may still wait. The company's system-card appendix identifies GPT-6 Astra as the model behind Dots, but a model description is not an independent test of the finished agent across real workflows. OpenAI explicitly cautions that a dot can make mistakes while following custom rules, so consequential actions still need proportionate human checks.
What users should verify next
A sensible first assignment is bounded: ask the dot to assemble a brief from a named folder or to flag calendar conflicts without sending messages. Inspect its source material, activity trail and proposed actions, then widen access only if the result is dependable. For organizations, test who can connect apps, what a dot can retain after disconnection, and how approval rules behave when a task changes direction. The launch establishes a new product surface; it does not yet establish a reliable outcome for every user or a general permission to act on their behalf.
Questions and answers
What is different about a dot?
OpenAI describes a dot as an ongoing agent that can continue work and return results between chats. A normal chat responds to a turn; a dot may have a standing goal, connected apps and scheduled tasks.
Can a dot use my computer automatically?
OpenAI says local computer access is optional and off by default. A user must connect the desktop app and allow access; it can later be revoked.
Is Dots available everywhere?
No. OpenAI's launch guide describes a gradual rollout with plan and regional restrictions. Availability should be checked in the current account rather than inferred from the announcement.