OpenAI launches Dots, always-on agents that work across connected apps

Why it matters

Dots shifts OpenAI's product pitch from conversational answers to persistent work across apps. Its value will depend on reliability and permission controls as much as on model capability.

OpenAI launched Dots on September 29th, a persistent AI agent that can work in the background across connected apps, as Sam Altman pitches a move from answering prompts to taking on ongoing work. In his X announcement, OpenAI's CEO described Dots as a way to use AI around the clock; the company's product announcement says each agent gets its own cloud computer and browser.

The launch puts Altman back in a familiar role: turning a technical product into a claim about how people spend their time. Before co-founding OpenAI, he co-founded location-based social network Loopt, which was acquired by Green Dot in 2012, and later led Y Combinator. Dots extends that product ambition into a tougher proposition: software that must keep working after the user stops prompting it. OpenAI is asking customers to trust an agent with context from work apps, let it act in the background, and judge its value by completed tasks rather than answers in a chat window.

OpenAI says Dots are powered by GPT-6 Astra, can connect to more than 4,000 apps through plugins, and can be reached in ChatGPT, Slack and Microsoft Teams. The company says they can research, analyze data, draft documents and prepare software changes, carrying context across channels. Its examples include an agent tracking customer feedback, preparing bug fixes and presenting pull requests for review, and another revising launch materials when a product's scope changes. Those are product scenarios supplied by OpenAI, not independent evidence of how reliably Dots complete the work.

OpenAI is building approval into the workflow. Users choose which apps a Dot can access and can set rules requiring approval for specific actions or blocking them. OpenAI says background "proactive research" uses read-only access when the user is not actively working with the agent; actions that could affect accounts or share information are checked, and some sensitive actions remain with the user. The company also cautions that Dots can make mistakes and tells users to review consequential work. That combination of broad access and explicit review controls is central to the product: an agent that can see more of a user's work may be more useful, while each new permission raises the cost of a mistake.

Availability is narrower than the around-the-clock pitch might suggest. OpenAI says Dots are rolling out in eligible markets to ChatGPT Pro and Business Premium users, with the first Dot included at no extra cost. Enterprise, Education and Healthcare customers can try a beta if their workspace administrator enables it. The first Dot is set up through ChatGPT on desktop or a desktop browser; users can message it on mobile after setup. OpenAI says conversations with a Dot do not count toward ChatGPT usage limits, while tasks it starts or manages in Codex or ChatGPT Work do count toward those limits. The company says plans include an allowance for deeper work, but did not specify its size in the launch post.

The first version is a personal assistant attached to an existing ChatGPT subscription, while OpenAI's enterprise offering is still at the pilot stage. The company says it is testing specialist Dots with dedicated identities and access to company systems, and is working with Microsoft to integrate them with Agent 365. Those agents would take on bounded organizational jobs, a different deployment problem from giving one user an assistant that learns their habits. The announcement does not provide customer results or independent measures of task completion, reliability or time saved.

OpenAI is entering a field that Meta is also pursuing with Muse, while Axios reported that Apple and startups including Instinct are part of the wider race to build personal agents. The initial difference is distribution: OpenAI is starting with eligible paid ChatGPT users and work tools, while Meta has pushed Muse toward a broad consumer audience. OpenAI's own live demonstrations at DevDay included both successful moments and a lag when an employee tried to speak to her Dot, according to Axios. The stumble is a small but relevant reminder of the product standard: background agents need to be dependable, not just capable in a prepared example.

Altman built his early career around consumer software and startup investing before OpenAI made him a central figure in the AI industry. Dots makes a more operational bet: that ChatGPT can become a persistent layer between users and the apps where work happens. OpenAI is offering the first agent as part of certain subscriptions, but its launch materials leave the practical test to users: whether Dots can handle routine work accurately enough to earn broader permissions and a place in the workday.