OpenAI Dots are rolling out as always-on AI agents that can take on ongoing work through their own cloud computers. Announced at DevDay on September 29, 2026, the agents are powered by GPT-6 Astra and can use connected apps while users set boundaries for what they may do independently.

The launch moves OpenAI beyond answering a prompt or completing a single delegated task. A Dot can watch for relevant changes, pursue a longer goal and return work for review, but the initial rollout is limited to eligible paid plans and markets.

OpenAI's announcement identifies four parts of the launch:

• A dedicated cloud computer for each Dot
• Connections to apps through plugins
• User controls for actions and approvals
• A staged rollout across paid plans

Related Research

• Meta Launches Meta One AI Subscriptions Across Instagram, WhatsApp and Facebook
• Nvidia Launches Open Agent Safety Platform With OpenShell and Sentry

OpenAI Dots Begin a Staged ChatGPT Rollout

OpenAI's product announcement says users can create a primary Dot, name it and give it ongoing responsibilities. The company is beginning the rollout for Pro and Business Premium subscribers in eligible markets, with Enterprise, Edu and Healthcare access available as a beta when an administrator enables it.

The first Dot is included in a Pro or Business Premium subscription. OpenAI says those plans include an allowance for deeper work, with temporarily extended limits during the first month. The company has not published a universal allowance in the announcement, so the included capacity may depend on the plan and rollout stage.

People can talk to Dots through ChatGPT on desktop, web and mobile, and OpenAI says Slack and Microsoft Teams are also supported channels. Text messaging is planned. The ability to contact an agent across channels matters because the work can continue after the original chat ends.

The Associated Press reported that CEO Sam Altman introduced Dots during OpenAI's annual developer conference in San Francisco. The announcement was part of a larger DevDay release that also included GPT-6.1 Sol and a premium Ultrafast speed tier, but Dots is a distinct agent product.

How a Dot Uses Its Computer and Connected Apps

Each Dot operates on an OpenAI-provided cloud computer with its own browser. That gives the agent an environment for using software and carrying work forward without requiring a person's laptop to stay open. A user can inspect the Dot's computer while it works.

OpenAI says Dots can reach more than 4,000 apps through its plugin ecosystem, subject to the connections and permissions a user has granted. The number describes available integrations, not apps that every Dot can automatically read or change.

The company illustrates the design with work such as monitoring customer feedback, preparing code fixes, updating analysis when new evidence appears and drafting launch materials when a project changes. These are examples of intended workflows rather than independently measured performance claims.

A Dot may work on several projects and learn a user's preferences from feedback. OpenAI says its goal is for the agent to bring completed work or decisions back at useful moments, including before a person issues another direct request.

That operating model raises a practical distinction: a persistent agent needs clear objectives, access limits and review points. The usefulness of the product will depend on whether it can recognize the right signal, finish a task reliably and avoid creating new work through mistakes or unnecessary activity.

Permissions and Approval Controls for Dots

OpenAI says users can define what a Dot may do by itself, which actions require approval and which actions are off limits. Each Dot's cloud computer remains separate from the user's device unless the user explicitly connects that device and grants access.

The company's announcement also describes safeguards against malicious instructions and monitoring that may pause or stop an agent when a safety concern appears. For actions that affect accounts or share information, OpenAI says an automated review checks the proposed step against user instructions and safety requirements.

OpenAI offers an example in which an early tester's Dot prepared an invoice and sent it only after the user's approval. That distinction is central to the product: an agent can assemble work in advance while a consequential external action still waits for a human decision.

Permissions also determine the quality of the result. An agent with too little context may miss the point of a task; one with broad access creates a larger risk if it misunderstands a request or follows untrusted content. Users will need to review connected apps and action settings as duties expand.

Enterprise Dots and the Next Test for Agent Work

OpenAI previewed specialist Dots for organizations with their own identities, managed hardware and deeper connections to company systems. These are separate from the personal Dot being rolled out now. The company says it is working with Microsoft to bring enterprise governance for specialist Dots into Agent 365.

OpenAI is starting those specialist deployments through focused pilots, with organizations defining responsibilities, tools and approval processes. It has not announced general availability or a standard price for them.

The launch also puts OpenAI into direct competition with other companies building personal and workplace agents. AP identified Meta's Muse as a rival. The meaningful comparison will be based on completed tasks, reliability, user control and cost, rather than a claim that an agent can be active all day.

Near-term checkpoints include the pace of the rollout, the published usage limits, developer support for plugins and evidence that Dots can complete repeatable workflows under clear permissions. Those details will determine whether always-on agents become a dependable way to delegate work or remain a promising demonstration.