OpenAI introduced Dots on September 29, 2026, as always-on agents designed to continue working between conversations. OpenAI says Dots can use a cloud computer and connected apps while operating under user-set permissions and approvals.
What makes Dots different
A normal chat session waits for the next prompt. An always-on agent is designed around a goal that can continue after the initial conversation. That changes the workflow from answering a question to managing a task over time.
Where the cloud computer fits
OpenAI says Dots run on their own cloud computer. This lets the agent interact with software and websites without requiring every step to happen on the user's local machine.
Why permissions matter
Connected apps can make an agent more useful, but access control becomes important. Users should give an agent only the connections needed for its current task and review approval settings before allowing actions with external effects.
How teams can test Dots
- Start with a low-risk recurring task.
- Connect only the apps required.
- Set clear approval points.
- Review completed actions.
- Measure time saved and corrections needed.