OpenAI is taking its push into agentic artificial intelligence a major step further with the launch of Dots, a new type of AI agent designed to keep working on a user's behalf even after the conversation with ChatGPT has ended. Announced at OpenAI's DevDay 2026, Dots are designed around an "always-on" model in which users give an agent a goal, provide access to the applications it needs and establish boundaries, while the agent continues making progress between conversations. Rather than waiting for another prompt, a Dot can monitor an ongoing project, determine what needs to happen next and return to the user when something requires attention or approval.
The idea represents a significant shift from the traditional chatbot model. ChatGPT has historically been something users actively open and interact with, whether they are asking a question, generating content, analyzing information or writing code. Dots are intended to behave more like persistent digital coworkers. OpenAI says they can learn a user's preferences from feedback over time, maintain context and continue working toward a goal around the clock. The company describes them as an "extension" of the user rather than simply another chatbot interface.
Dots are powered by GPT-6 Astra and operate using their own cloud computers and browsers. That means an agent can interact with websites, applications and other connected tools without necessarily operating directly on the user's personal computer. OpenAI says this separation is an important part of the system's safety design, giving the Dot its own environment in which it can perform work while keeping the user's computer isolated unless the user explicitly chooses to connect it.
The agents can also connect to a large ecosystem of software. OpenAI says its plugin ecosystem gives Dots access to more than 4,000 apps, allowing them to work across services rather than being restricted to the ChatGPT interface. The company's examples include Google Calendar, Google Drive, GitHub and other productivity and development tools. Dots can also be accessed through ChatGPT as well as workplace platforms including Slack and Microsoft Teams, allowing them to remain connected to projects where work is already happening.
That connectivity is what makes Dots substantially different from an AI assistant that simply generates a response. A user could give a Dot responsibility for monitoring customer feedback, for example, and the agent could continuously review incoming information, identify recurring issues, investigate the relevant code and prepare changes for review. OpenAI says its internal examples include Dots investigating bugs as they appear in Slack, turning design changes into working applications and updating project materials as new information arrives.
The same concept can be applied to more conventional office work. A product manager could give a Dot responsibility for maintaining launch materials, while a sales employee could use one to keep proposals updated as customer requirements change. Researchers could have a Dot rerun analyses when new data becomes available, while developers could use one to monitor customer feedback and prepare tested software improvements. The important difference is that the agent continues working on those responsibilities rather than requiring the user to restart the process every time something changes.
OpenAI is also pitching Dots as useful for personal tasks. In one example, a Dot notices that a user's calendar indicates they are working through dinner and provides food-delivery options with pricing and delivery times. The user chooses an option and approves the order, after which the Dot can follow through with the task. That kind of interaction illustrates the company's broader vision: the AI does the research and preparation in the background while the human remains responsible for decisions that require approval.
The phrase "always watching" may sound unsettling, but OpenAI's implementation is more specifically about persistent task monitoring than unrestricted surveillance. Dots are given access to particular applications and information sources selected by the user, and their ability to act can be restricted through permissions and custom rules. OpenAI says users can decide which apps a Dot can access, what it should focus on and which actions it is allowed to take.
Those controls are particularly important because a truly autonomous AI agent can potentially cause problems if it has too much freedom. OpenAI says Dots include built-in safeguards, permission controls and action-review systems designed to determine when an action is safe to execute automatically and when the user needs to approve it. Actions that could affect accounts or share information can be subjected to additional review, while particularly sensitive actions such as changing a password remain under explicit user control.
OpenAI also says background work is subject to additional restrictions. According to the company's announcement, Dots operating proactively in the background are limited to read-only tools in certain circumstances, preventing them from independently sending messages, modifying application content or controlling the user's computer. This is intended to create a separation between an agent's ability to observe information and its ability to take consequential actions.
The security model is particularly important because agentic AI has already generated concerns about what autonomous systems can do when given access to real-world tools. OpenAI's launch comes amid increased scrutiny of AI agents following incidents in which autonomous systems reportedly behaved in unexpected ways, including incidents involving external websites and data. Reuters reported that OpenAI emphasized custom rules and approval requirements for Dots as part of its response to those broader concerns.
OpenAI is therefore attempting to solve two problems simultaneously: making AI substantially more autonomous while keeping the human in control. The company wants Dots to be capable of deciding how to pursue a goal without requiring step-by-step instructions, but it also wants users to establish the boundaries within which those decisions can be made. In practice, that means users are increasingly being asked to think less about individual prompts and more about delegating responsibilities.
That distinction could become one of the most important changes in the way people interact with AI. With a conventional chatbot, the basic interaction is "ask and answer." With an agent, it becomes "assign and supervise." The user describes the outcome they want, gives the AI the resources it needs and then monitors the work as it progresses. Dots take that concept and make it persistent, allowing the relationship with the AI to continue even when the user is no longer actively chatting with it.
OpenAI is also positioning Dots as part of a larger workplace strategy. The company's launch coincides with the introduction of ChatGPT Space, a shared environment where people and AI agents can work with common files, projects and context. Dots can potentially become participants in those collaborative environments, helping teams keep documents updated, track project tasks, investigate issues and prepare work for human review.
The timing puts OpenAI directly into competition with other companies pursuing persistent AI agents. Meta has been developing its own agentic products, including Muse, while a growing number of startups are building autonomous systems designed to perform tasks rather than simply answer questions. Reuters described Dots as another front in the competition to move beyond chatbots and sell AI that can actually perform work on behalf of users.
The competitive stakes are especially high in the enterprise market. Businesses have spent years purchasing software designed to help employees manage sales, customer service, development, finance and project management. Agentic AI introduces the possibility that an AI system could operate across many of those applications instead of requiring workers to manually move information between them. A Dot could potentially read a customer request in one application, analyze information in another, update a document and prepare a response without the employee having to coordinate each step.
That could eventually change the role of conventional software interfaces. Instead of employees opening multiple applications and navigating each one individually, they could increasingly delegate tasks to an AI agent that works across those systems. The applications would still exist underneath, but the AI could become the layer through which people interact with them.
OpenAI's own examples point toward that future. A developer could ask a Dot to monitor customer feedback, identify a recurring complaint, investigate the problem, build a fix and test it. A content creator could provide an interview transcript and have the Dot identify clips, create show notes and prepare social posts. A sales employee could give the agent changing customer requirements and have it continually update a proof of concept or proposal.
The challenge is that autonomy also creates a new category of responsibility. If an AI agent makes a mistake while working continuously, the mistake could potentially compound over time. An incorrect assumption could influence multiple documents, an inaccurate analysis could be repeated as new information arrives or an inappropriate action could affect an external account. The longer an agent operates without supervision, the more important its ability to recognize uncertainty and request human intervention becomes.
OpenAI's response is to emphasize control rather than simply maximizing autonomy. The company says users can define custom rules that specify what Dots can do, what actions require approval and what actions are blocked entirely. The platform also includes automated reviews intended to check potentially consequential actions against those rules and OpenAI's safety requirements.
Privacy is another major consideration. A Dot that is genuinely useful needs access to meaningful context, and that could include calendars, documents, repositories, workplace conversations and other information. OpenAI says users choose which apps to connect and that business customers' data is not used for model training by default. Personal-plan users can also control whether their conversations and Dot activity are used for training.
OpenAI is initially limiting availability. Dots are rolling out to eligible Pro and Business Premium users aged 18 and older in supported markets, with Enterprise access in beta and controlled by workspace administrators. The company's release notes say Pro access excludes the European Economic Area, Switzerland and the United Kingdom at launch.
The company is also giving users a way to interact with their Dots across multiple environments. Users can communicate with them through ChatGPT on web, desktop and mobile, while workplace integrations extend the agents into Slack and Microsoft Teams. OpenAI says additional communication options are also being explored, including text-based interactions.
The launch comes as OpenAI attempts to turn ChatGPT from an application people visit into an AI platform that can continuously participate in their work. That is a much more ambitious vision than simply making the chatbot smarter. If Dots work as intended, users could eventually have AI agents responsible for entire categories of recurring work, with the human stepping in primarily when judgment, approval or creative direction is needed.
That also explains why OpenAI describes Dots as "an extension of you." The company is not simply selling another AI assistant. It is proposing a model in which an AI system learns how an individual or organization prefers to work and then attempts to reproduce those preferences while handling ongoing responsibilities. The more feedback the Dot receives, the more OpenAI says it can adapt to the user's preferences and expectations.
Whether that vision succeeds will depend heavily on trust. People may be comfortable asking an AI to draft an email or summarize a document, but allowing an AI to monitor work continuously and take actions independently is a substantially different proposition. Users will need confidence that the agent understands its boundaries, protects sensitive information and knows when to stop and ask for permission.
For OpenAI, Dots represent a clear attempt to move ChatGPT into that next phase. The chatbot is becoming less about answering individual prompts and more about maintaining an ongoing relationship with the user's projects and responsibilities. With its own cloud computer, access to thousands of apps, persistent context and the ability to work between conversations, a Dot is designed to keep moving even when the user walks away.
The result is a new kind of ChatGPT experience in which the AI is no longer simply waiting for the next question. Instead, users can hand over a goal, establish the rules and allow the agent to keep working toward the outcome. That shift—from AI that responds to AI that persists—could become one of the defining changes in the next generation of ChatGPT.