OpenAI has launched dots, always-on AI agents that can work across connected apps and continue tasks between conversations. Announced at DevDay on September 29, 2026, dots are rolling out first to eligible ChatGPT Pro and Business Premium users, with an admin-enabled Enterprise beta. The important shift for businesses is not a chat interface; it is the decision to give an agent continuing access to tools, information and, sometimes, the authority to act.
Key takeaways
- OpenAI dots run on GPT-6 Astra and use their own cloud computers; a user’s local computer is optional.
- OpenAI says the first dot is included with eligible Pro or Business Premium plans, but deeper work has an allowance and delegated Codex work uses normal limits.
- Connected-app permissions, custom rules, an activity view and action review determine what a dot can read and do.
- For Indian teams, the practical test is whether an agent can finish a bounded task with a clear approval trail, not whether its launch demo looks autonomous.
What OpenAI dots actually change
A conventional chatbot waits for a prompt and returns an answer. OpenAI describes dots as AI agents that can retain a responsibility, inspect the next step, and keep working after the user leaves a conversation. The company says each dot has a cloud computer and browser and can work with apps that the user chooses to connect. OpenAI lists examples including following a project, preparing a document when requirements change, and bringing completed work back for review. These are company examples, not evidence that every dot can perform every such job reliably.
At its September 29 launch, OpenAI said dots are powered by GPT-6 Astra. The company also says its plugin ecosystem can connect to more than 4,000 apps. That is a vendor-stated integration count; it does not mean every app or capability is available to every user or region. WIRED’s launch report independently described the multistep task and connected-app approach. Axios, reporting from DevDay, noted that live demos included a voice-response lag, a useful reminder that an announcement is not an independent reliability test.
Dots also arrive in a crowded market for persistent personal agents. Meta launched Muse the week before OpenAI’s announcement, and our earlier coverage explained Muse’s connected small-business workflows. The two products may invite comparison, but the meaningful question for an organisation is narrower: which accounts can the agent reach, what may it change, and where must a human approve a step? Those settings shape the risk and usefulness of both services more than their names or avatars.
AI agents make the permission boundary the product
OpenAI’s launch material emphasises that the dot’s computer is separate from a user’s own device. The user can connect a local machine deliberately, but the default separation matters: it limits the starting surface for mistakes and makes the grant of extra access a visible decision. In an Indian small business, for example, there is a material difference between asking an agent to read a sales report and giving it access to modify the accounting system that produced that report.
The company’s safety explanation says users choose connected accounts and can set Custom Rules for actions. It describes an additional auto-review step for account-changing or information-sharing actions, and monitoring that can pause suspicious work. OpenAI also says background proactive research is limited to read-only tools; any follow-up that changes something is subject to ordinary checks. Those are OpenAI’s descriptions of its controls, not a guarantee that errors or malicious instructions cannot get through.
That distinction is more than legal caution. A long-running agent may encounter emails, documents and websites created by other people. Those pages can carry instructions that try to redirect the agent, a problem known as prompt injection. The company’s safety post explicitly names this risk and says its response combines tool limits, action checks and monitoring. WIRED also warned that connecting personal data creates practical privacy and security questions, even with safeguards. A team should therefore review the permissions for each connection and the result of each important action, rather than treating the AI’s stated objective as a substitute for controls.
There is a second boundary: who is responsible when the agent acts. OpenAI’s product page says some tasks are returned for approval; its help article says users can define what a dot can do on its own. A company can use that to assign low-risk preparation and keep payment, account, publication or client communication decisions with a person. OpenAI says particularly sensitive actions, such as a password change, stay with the user. The design is closer to delegated work with checkpoints than to a blank cheque.
Who can use OpenAI dots now?
Availability is narrower than the broad launch language can suggest. OpenAI’s current help article says dots are rolling out to Pro users outside the European Economic Area, Switzerland and the UK, and to Business Premium users in supported ChatGPT regions. Enterprise, Edu and Healthcare customers can try a beta when a workspace administrator enables it; the feature is initially off by default. The rollout is gradual, so a qualifying subscription does not necessarily produce immediate access.
That means eligible users in India may be able to join the rollout, but this article does not assume that every Indian account already has dots. OpenAI says users create their first dot on the ChatGPT desktop app or desktop web. The help page says users cannot currently create one on a mobile device, although they can later interact with it in the mobile app when mobile access is available. Text messaging is also a limited US beta at launch rather than a general India feature.
OpenAI says the first dot is included in Pro or Business Premium at no extra cost. It is careful to distinguish ordinary conversations with a dot from an allowance for deeper work, and says tasks handed to Codex or ChatGPT Work count against those products’ normal usage limits. Neither the product page nor the help page provides a simple universal figure for the amount of deep work a first dot can do. Buyers should check the live plan details and their account’s limits rather than assuming that “always on” means unlimited execution.
| Plan | Availability | Important condition |
|---|---|---|
| Pro | Gradual rollout in eligible markets | Excluded at launch in EEA, Switzerland and UK |
| Business Premium | Supported ChatGPT regions | Gradual rollout |
| Enterprise, Edu, Healthcare | Beta | Workspace admin must enable it |
Why the India business use case needs a tighter test
For a founder or operations lead, the persuasive use case is rarely “an agent can do everything.” It is a repeated task that has a measurable completion standard: reconcile a changed product brief with website copy, assemble evidence for a sales proposal, or draft a response from approved documents. OpenAI’s examples include several such workflows, but they are illustrative claims by the vendor. An organisation should test one narrow process with its own data, permissions and review requirements before expanding the scope.
That is especially relevant when a task touches customers. The agent might assemble a proposed email or update an internal slide, but sending a binding quote or publishing a live claim is a separate action with a different cost of error. A sensible pilot records the input sources, what the dot attempted, what changed, and who approved release. OpenAI says Activity View makes ongoing and delegated work visible; the practical question is whether that view gives the organisation a sufficient audit trail for its process.
Agents also intensify the ordinary problem of access sprawl. If a person connects an inbox, files, calendar and business software to one assistant, the assistant can potentially infer more than any one app reveals. OpenAI’s controls let users choose connections and manage app permissions. In an organisation, however, individual choices must still fit company data policy. The same governance issue appears in our reporting on specialist agent-governance tools and in NVIDIA’s agent-safety control layer. The question is not simply whether a model refuses a bad prompt; it is whether a team can observe and restrict the actions available to it.
OpenAI is also previewing specialist dots for enterprises, with distinct identities and IT-provisioned hardware, according to its announcement. These are pilots, not a general product every business can buy today. Its proposed integration with Microsoft Agent 365 is likewise described as work in progress. A small Indian firm choosing a workflow this month should evaluate the dot it can actually access, rather than capabilities announced for a future managed deployment.
Dots versus a one-off ChatGPT task
A one-off prompt is useful when the input is stable and the output can be judged immediately. An ongoing agent is useful when the information changes and someone otherwise has to remember to check it. That shift creates a recurring duty to inspect sources and decide when to intervene. It also means a mistake can persist or recur if the initial objective is vague. The right measure is not how long the agent runs; it is whether it makes correct progress within the authorised boundary.
OpenAI describes dots as carrying context across ChatGPT, Slack and Microsoft Teams. That can reduce repetition for a project moving between channels. It also increases the value of setting one clear owner for the task and knowing which channel contains the authoritative decision. OpenAI says a dot can bring work and decisions back to the user. Teams should make that return path explicit: who signs off a draft, who checks the underlying data, and when should the agent stop rather than guess?
Independent reporters have focused on the competitive story: Axios framed dots as OpenAI’s response to Meta’s Muse, while WIRED reported the privacy trade-offs of always-on assistance. The Guardian separately covered the launch. Our reading is that the durable business test is less about which company introduced a persistent mascot first and more about whether users can delegate a real job without losing track of authority, evidence and cost.
What remains unproven after DevDay
OpenAI’s materials explain how dots are intended to work, but they do not publish a universal completion rate for real-world business tasks, a common latency benchmark across apps, or a fixed deep-work quota for every plan. The launch should therefore be read as product availability and design intent, not proof that every workflow is ready for unattended use. Axios’s report of a lag in one live demo is an observation about a demo, not a general performance measurement.
Availability may also change quickly. OpenAI’s help page says the rollout is gradual and some features might reach accounts later. Vendor claims about supported integrations and planned enterprise features may shift as the product develops. This article is based on materials live on September 30, 2026; readers planning a purchase or deployment should confirm the current entitlement and terms in their own account.
The practical conclusion is straightforward: OpenAI dots turn AI agents into continuing workers inside ChatGPT, but the value of that shift depends on access design. Start with a small task whose source material, permitted actions and human approval point are explicit. If the dot completes it accurately and leaves a clear trail, widen the scope carefully. If it cannot, the attractive “always-on” label adds complexity without dependable output.
Frequently asked questions about OpenAI dots
Are OpenAI dots available in India?
OpenAI says eligible Pro users outside the EEA, Switzerland and UK and Business Premium users in supported regions are part of a gradual rollout. India is not named among the excluded Pro markets, but access may take days and depends on the account and plan. Check the ChatGPT desktop or web interface for your own availability.
Can a dot use my own computer automatically?
No. OpenAI says each dot has a separate cloud computer. Connecting a personal computer is an optional step that the user must allow. The connected apps and permissions are also chosen by the user or controlled by a workspace administrator.
Are OpenAI dots free or unlimited?
OpenAI says the first dot is included with eligible Pro and Business Premium plans at no extra cost. It also says deeper work has an allowance and tasks delegated to Codex or ChatGPT Work use those products’ normal limits. “Always on” does not establish unlimited work.
What is the difference between dots and ordinary AI agents?
OpenAI’s dot is a specific consumer and business product built on GPT-6 Astra. It is designed to hold an ongoing responsibility, use a cloud computer and permitted connected apps, and keep progressing between conversations. “AI agents” is the wider category of software that can plan or carry out tasks using tools; other products may have different controls, pricing and capabilities.
Get the day’s top stories in your inbox
One concise email. No spam, unsubscribe anytime.
