OpenAI has launched Dots, an always-on AI assistant designed to perform work across connected applications in the background. The product arrives weeks after Meta introduced its own customizable agent, Muse, making the competitive shift clear: major AI platforms are trying to turn conversational assistants into systems that can execute multi-step work.
Dots is rolling out to ChatGPT Pro, Business Premium and Enterprise customers. It is powered by OpenAI’s GPT-6 Astra model and operates through its own cloud computer, with access to a web browser and more than 4,000 supported apps, according to OpenAI.
From chat interface to delegated work
A Dot can be assigned a task through a messaging-style interface, then report progress, ask questions and surface results while it works. Users can also call a Dot by voice through ChatGPT on web, desktop or mobile. Integrations with Slack and Microsoft Teams are intended to carry context across workplace conversations and devices.
OpenAI’s examples point to the kind of workflow it wants to own. A developer might ask a Dot to review customer feedback, build and test product updates, then present the changes in a video. A creator might use one to analyze an interview transcript, identify clips, produce show notes and draft social posts.
That is a more consequential proposition than a chatbot that generates a first draft. It places the agent inside a chain of activities that can involve customer data, source code, publishing systems and collaboration tools. For operators, the relevant test is therefore not whether a Dot can produce an impressive demo, but whether it reduces handoffs and cycle time without creating unacceptable review overhead.
The governance layer will determine enterprise adoption
OpenAI says Dots include built-in rules governing when they may act independently. Customers can add rules to prohibit actions or require permission, and an “auto-review” capability is meant to check actions against those rules and safety requirements. Some actions, such as changing a password, can be handed back to a user.
Users can also access a Dot’s cloud computer to follow its activity. That visibility matters because agentic products are judged differently from standard software assistants: an inaccurate answer is one problem; an incorrect external action can be a security, financial or reputational incident.
The timing makes those controls particularly important. Meta’s Muse has drawn attention for its broad ability to browse, purchase, generate documents and track goals, alongside reported concerns about unsafe data sharing. OpenAI is positioning permissions, review and user-defined boundaries as core features rather than optional enterprise add-ons.
Teams evaluating Dots should still establish their own guardrails. Start with bounded, reversible workflows; define which systems and data the agent may access; require approval for external communications or irreversible changes; and identify a human owner for each deployed agent. Logging and review processes will be as important as prompt design.
A path toward an agent workforce
At launch, each user can create one Dot. OpenAI says it plans to support multiple simultaneous agents and to let customers increase each agent’s speed or monthly work capacity. It is also testing “specialist” Dots that companies can configure for specific organizational roles.
That roadmap shifts the unit of value from an individual assistant subscription to managed capacity: multiple narrow agents, each working under defined access and escalation policies. OpenAI also says Dot conversations will not count against ChatGPT usage limits, removing one practical constraint on continuous work.
A new collaborative workspace, ChatGPT Space, lets colleagues work with a Dot together. This could make agents more useful for shared operational processes, but it also raises familiar questions over access controls, accountability and who can authorize a task.
What to watch next
The immediate question is whether Dots can reliably complete useful work in real production environments—not merely navigate apps, but handle exceptions, preserve context and know when to stop. The next indicators will be the breadth and quality of integrations, the controls available to enterprise administrators, and whether specialist Dots can be deployed with clear audit trails.
For builders and business leaders, Dots is a sign that the AI-agent race is moving beyond model quality toward workflow execution, governance and distribution inside the tools people already use.
