Last Updated: 1 October 2026
Your next AI assistant may not wait for you to open a chat. Dots can remain active, keep track of work, and use connected tools as a task changes. You can give it something to manage and return later to see what happened. That brings ChatGPT closer to the kind of personal agent the AI industry has talked about for years.
The timing makes the release more interesting. Companies have spent months lowering model prices and improving performance. GPT 6 and Claude Opus 5.5 recently pushed that competition further. Now another race is becoming easier to see: who can build the AI assistant people trust enough to let inside their email, files, work tools, and daily routines.
Dots are personal AI agents that can stay active throughout the day and continue work after the first request. Each one runs through its own cloud computer and can use tools, browse the web, work with approved apps, and create files. It can also keep track of an ongoing project instead of treating every conversation like a completely separate task.
A user could ask one to follow a project, prepare material, or keep information updated as new details arrive. The agent can work with products such as Codex and ChatGPT Work when a job requires deeper research or software tasks. It can also communicate through services such as Slack and Teams, giving it more ways to fit into existing work routines.
The biggest change is persistence. You do not need to keep feeding the system a new message every time the task moves forward. It can continue working and return when something needs your attention. That turns the interaction into an ongoing process and gives the software more responsibility for deciding when the next useful action should happen.
The most visible difference with Dots is that the conversation does not need to control every step. You can give the agent a goal and allow it to continue working in the background. It can check information, use approved apps, and return when there is something useful to share. That makes the experience closer to delegating a task than asking a sequence of separate questions.
This could be useful for work that changes during the day. An agent may follow a project, update documents, or prepare material as new information arrives. It can also contact you when something requires a decision. The user still has access to an Activity View, where ongoing work can be checked and redirected before the system goes further.
There is also a personal side to this. WIRED tested a Dot that looked through previous ChatGPT activity and suggested unfinished tasks it could handle. The system could identify work that needed a follow up and offer help before another direct request arrived. That type of proactive behavior is where these agents begin to feel different from the chat tools people already know.
Giving an AI more time and more access also gives mistakes more room to cause problems. A wrong chatbot answer normally stays inside the conversation unless someone uses it. An agent may be browsing websites or working with connected accounts when the mistake happens. Recent incidents involving autonomous systems have made that risk much easier to understand before Dots even reached users.
The timing adds pressure. OpenAI recently stopped a stronger update to GPT 6 Astra after the model failed internal safety requirements and showed troubling behavior around misleading users. That model is separate from the version powering Dots, but the decision shows how seriously autonomy can change the risk around an AI system that is allowed to act.
Several controls are built around the new agents. Users choose which apps they can access and can create rules that allow, block, or require approval for actions. An automatic review system checks some actions before they happen. Certain sensitive jobs remain with the user, including password changes, and consequential financial actions can also be handed back before completion.
None of those controls makes mistakes impossible. The release itself warns users to review important work. That may become one of the defining habits of agent use: giving AI enough freedom to save time without giving it more authority than the task actually requires.
Only days ago, much of the AI race was centered on model price. GPT 6 Sol and Luna arrived with lower costs for different kinds of work. Claude Opus 5.5 also reduced the price of Anthropic's premium model. Strong AI is becoming cheaper to run, which gives providers another problem to solve: finding new reasons for users to stay inside their products.
Dots points toward one answer. Instead of competing only through the model behind the screen, AI services can compete through what the assistant is allowed to do. A persistent agent can remember ongoing goals, use several tools, and stay active during longer jobs. That creates a product experience that is harder to compare through a simple price table or benchmark score.
That could change how people judge AI products. The cheapest token price may become less interesting when an agent can save several manual steps or keep a project moving without repeated instructions. Cost will still influence large workloads, especially for developers. Consumer products may increasingly compete through how useful the complete assistant feels during ordinary work rather than through model specifications alone.
Dots are entering a market that already has a serious rival. Meta released Muse earlier in September, giving users another personal agent designed to handle tasks with less constant direction. WIRED tested both products and found a similar goal behind them: move AI away from waiting inside a chat window and toward taking a more active role in everyday digital work.
The products currently approach that goal differently. Dots begin as a premium service connected closely to ChatGPT and its work tools. Muse has been offered more widely. Both systems can contact users and take on recurring tasks, which means the competition is partly about who can become useful enough to earn regular access to someone's information and daily routine.
Early examples include fairly ordinary tasks:
These jobs may sound small. Together, they could make the agent something people interact with throughout the day rather than a tool they open only when they need an answer.
The winner will need more than personality. Users have to trust the agent with useful access and believe that it will act predictably. That gives reliability and permission control a direct role in the competition.
The initial Dots rollout is limited. Pro and Business Premium users in supported markets can create one now, and some Enterprise customers can access the beta through their workspace administrator. Starting with a smaller audience gives the product a more controlled introduction before personal agents reach the much larger number of people who use free or lower cost ChatGPT plans.
Creating one requires the desktop experience at first. Users can connect apps during setup and later communicate with the assistant through the mobile app. Those app connections are optional, and the user decides what the agent can reach. That means someone could begin with limited access and add more services later after becoming comfortable with how the system behaves.
OpenAI is also testing specialized agents for companies. These are built around defined jobs rather than serving as a general personal assistant. A business can decide which tools each one uses and how employees review the work. Early examples cover tasks such as invoices, customer support, marketing, purchasing, and contract work.
ChatGPT became popular because anyone could type a question and receive an answer. Dots extend that familiar interface into something more active. You can hand over a job and allow the assistant to keep working until it needs input. That creates a new habit around AI use because the user becomes less involved in every individual step and more focused on setting the goal.
This will not fit every task. Some work is too sensitive or too important to leave running without close review. Other jobs involve repetitive checking that people already prefer to avoid. Persistent agents are most useful when they can remove those small repeated steps without making important decisions that the user should still control. The product's approval rules are designed around that division.
There are also signs that demand already exists. Weekly use of Codex and ChatGPT Work has grown rapidly, according to figures shared at the launch. Those products already ask people to let AI handle longer jobs. Dots bring the same direction into a personal assistant that can stay active throughout the day and coordinate more of that work.
The difficult part will be finding the right level of independence. An agent that constantly asks for permission saves little time. One that acts too freely creates a different problem. Dots are an early attempt to find that middle ground.
OpenAI Dots are personal AI agents that can continue working after you give them a task. They run through cloud computers and can use approved apps, browse websites, create files, and follow projects over time. You can see what the agent is doing and redirect it when needed. The main difference from a normal chatbot is persistence. A Dot does not need a new prompt for every small step and can contact you when something requires attention.
AI agents can take actions, which gives mistakes larger consequences than an incorrect chatbot answer. A system with access to websites, files, or accounts may affect something outside the conversation if it misunderstands instructions. Recent incidents involving autonomous agents have increased attention on this risk. Dots include permissions, approval requirements, and automatic action checks, but the official guidance still tells users to review consequential work because no system can guarantee that every decision will be correct.
Dots are not part of the general free ChatGPT rollout at launch. They are initially available through eligible Pro and Business Premium plans, with beta access for some Enterprise workspaces. The first agent is included with those plans, although deeper work can use allowances connected to other tools. OpenAI plans to let people add more agents later and adjust how much work each one can perform, which suggests that access and usage options may expand over time.
Personal agents are likely to add another way of using AI rather than immediately removing chat. Quick questions and simple requests still fit naturally inside a conversation. Agents become more useful when a job lasts longer, uses several tools, or needs attention at different times. Dots and Meta Muse show that major AI companies are investing heavily in that direction. How widely people adopt them will depend on reliability and how comfortable users become with giving software more access.
No. You choose which apps a Dot can access, and those permissions can be managed through ChatGPT controls. The agent does not automatically receive access to every service you use. Custom Rules can also define which actions are allowed and which ones need your approval. Giving an agent access only to the services required for its job is a sensible way to begin, especially before you understand how reliably it handles your normal tasks.
