OpenAI’s Dots Put an AI Agent to Work Around the Clock

OpenAI launched Dots on September 29, 2026, with a pitch that moves beyond asking an AI chatbot for an answer: give an agent a goal, and let it keep working. A dot has its own cloud computer and browser, can use connected apps, and can return with progress or decisions that need its owner’s attention. OpenAI describes the agents as always on, though what they can access and do depends on permissions.

The immediate appeal is continuity. Instead of starting a fresh conversation each time a project changes, a user can keep one dot informed about a goal and assign it several strands of work. OpenAI says dots can learn from feedback, carry context between conversations and continue working while their owner is away.

What can a dot do?

OpenAI’s examples span software development, research, sales and content production. A dot might watch customer feedback, prepare and test a small software fix, then present the result for review. It might revise launch materials when a product changes, update an analysis as new data arrives, or prepare show notes and social posts from an interview transcript. The company also describes an early tester whose dot prepared a forgotten invoice and sent it after receiving approval.

Users can inspect a dot’s cloud computer and communicate with it through ChatGPT on desktop, web or mobile. Slack and Microsoft Teams are also supported where enabled; text messaging is planned. A user can separately permit a dot to connect to a laptop, rather than giving it access to that device by default. OpenAI says its plugin ecosystem can connect dots to more than 4,000 apps, but that figure does not mean a new dot automatically has access to them.

There is an important distinction within the always-on promise. OpenAI says that when a user is not actively working with a dot, its background “proactive research” uses connected apps through read-only tools. In that mode, it cannot send messages, alter app content or control a browser or computer. Other assigned work may involve actions, subject to the available permissions, built-in rules and approval checks. OpenAI says some sensitive steps, including changing a password, remain the user’s responsibility.

Who gets access?

Dots began rolling out in eligible markets to ChatGPT Pro and Business Premium subscribers on September 29. Enterprise users, including those in Edu and Healthcare workspaces, can try a beta when an administrator enables it. OpenAI says the first dot is included with a Pro or Business Premium plan at no additional charge, but deeper work has a plan allowance; expanded limits apply for the first month after launch. Conversations with a dot do not count toward ChatGPT usage limits, while tasks started in Codex or ChatGPT Work count toward those products’ limits as usual.

Users set up their first dot in the ChatGPT desktop app or a desktop browser, after which they can message it in the mobile app. OpenAI says access will expand, but has not given a date for a general rollout. In Enterprise workspaces, dots are off by default, and administrators can separately control features such as messaging integrations and local computer access.

How does it compare with Muse?

Meta introduced Muse on September 8 as another agent that continues working after a person closes the app. Muse also uses a dedicated cloud computer and browser, connects to services its user authorizes, and seeks approval for sensitive actions such as sending an email or making a purchase. Meta offers it through a dedicated app, WhatsApp and the web. It says Muse is free for most tasks, with subscriptions for people who want to do more.

The clearest distinction at launch is how the two companies are reaching users. OpenAI is starting Dots with specified paid plans and an administrator-enabled enterprise beta. Meta has presented Muse as a broader consumer product and has also added tools for small businesses. Both companies want their agents to remember context and advance longer-term goals; neither company’s launch description establishes that its agent will reliably complete every task without human intervention.

The question behind “always on”

An agent that can monitor connected information and act later demands a different kind of trust from a chatbot that responds only when prompted. Users need to know what an agent can see, how long it retains useful context, which actions require approval and how to check work completed while they were away. In workplaces, administrators must also decide which accounts and devices an agent should be allowed to use.

OpenAI offers app controls, an activity view, action reviews and restrictions on background research. Those safeguards make the boundaries more concrete, but they do not eliminate mistakes. The practical test for Dots—and Muse—is whether people can delegate enough work to save time while still understanding, correcting and stopping what their agents do.