OpenAI launches dots agents for ongoing work in ChatGPT

OpenAI announced dots, AI assistants designed to keep working across projects and connected apps, at its DevDay conference in San Francisco on September 29, 2026. The company says the first dot is rolling out in ChatGPT for Pro and Business Premium users in eligible markets, offering a way to delegate continuing work rather than issue a separate prompt for every step.

The launch moves OpenAI’s agent ambitions into a product that can use a cloud computer, browser and connected apps on a user’s behalf. OpenAI describes those capabilities and safeguards in its announcement; neither its examples nor the conference demonstration establishes how reliably dots complete real-world tasks.

What OpenAI says dots can do

OpenAI says dots run on GPT-6 Astra and can work on several projects between conversations with a user. Each dot has its own cloud computer and browser. Users can open that computer to inspect work in progress, and can separately permit a dot to connect to and use their own laptop. That distinction matters because the company says the user’s computer stays separate unless they choose to connect it.

Through OpenAI’s plugin ecosystem, dots can connect to more than 4,000 apps, the company says. It describes access through ChatGPT, Slack and Teams, with texting planned for later. Users can message or call a dot in ChatGPT on desktop, web and mobile, although OpenAI says initial setup takes place on a desktop browser or in the desktop app before mobile messaging becomes available.

The company also recounts an early tester whose dot noticed an invoice had not been sent, prepared it and sent it after the tester approved. OpenAI does not provide independent verification of that anecdote. CBS News reported that a DevDay stage demonstration showed workplace uses through Slack and Teams, including brainstorming, calendar updates and reviewing customer feedback.

Who can use dots at launch

OpenAI says the first dot is included at no extra cost with Pro and Business Premium plans in eligible markets. Enterprise users, including those in Edu and Healthcare workspaces, can try a beta when an administrator enables it. The announcement does not list eligible markets, so it does not establish availability in any particular reader’s location.

The company says plan allowances cover deeper work and that extended limits apply during the first month after launch, but it does not quantify those limits. Conversations with a dot do not count toward ChatGPT usage limits, according to OpenAI. Tasks a dot starts or manages in Codex or ChatGPT Work still count toward the applicable usage limits.

Permissions and work done in the background

OpenAI says users choose which apps a dot may access and can manage permissions through ChatGPT controls. Custom Rules can allow actions, require approval or block them, while certain sensitive tasks, such as changing a password, remain with the user. The company says people can follow a dot’s progress in Activity View and redirect it as work proceeds.

When a user is not actively working with a dot, OpenAI says it may conduct ‘proactive research’ using connected apps. The company describes the tools available for that background work as read-only: they cannot send messages, change app content or control the user’s browser or computer. Actions that could affect accounts or share information go through an automatic review against instructions, rules and safety requirements, OpenAI says.

OpenAI also says its systems monitor for malicious instructions and potentially harmful behavior and may pause or stop work. It says Business, Enterprise and Edu workspace content is not used to improve its models by default. Personal-plan users can control whether their conversations and work are used for that purpose. These are the company’s descriptions of its controls, not an independent assessment of how they perform.

What follows the initial rollout

OpenAI is separately beginning focused pilots of specialist dots for organizations. Unlike a personal dot working for one user, a specialist dot would have its own identity and defined responsibilities. OpenAI says its engineering teams will work with participating organizations to set each dot’s duties, tools and review process. It is also working with Microsoft on integration with Agent 365 governance and security controls.

The dots launch comes alongside a separate OpenAI model-release issue covered in NewsJaws’ earlier DevDay report. That report described a model release being held over security concerns; the hold was not a postponement of the dots announcement. CBS News reported that OpenAI positioned dots for professional work at DevDay, while also reporting safety concerns about agents more broadly.

The questions for users are practical: whether dots are available in their market, how much work their plan permits and how often a task requires correction or approval. OpenAI has not published quantified task-completion or error rates for dots in its announcement. Its stated controls explain how users are meant to supervise an assistant that keeps working between interactions; they do not establish its real-world reliability.