OpenAI is rolling out Dots, persistent agents that can work on a cloud computer and use connected apps. Eligible users get one primary Dot at launch, not a team of agents that can freely act across every connected service.
Dots moves ChatGPT toward work that continues beyond a single conversation. OpenAI says a Dot can pursue a task over time and research information in connected apps. It also describes rules for actions that affect accounts. Texting, multiple Dots and broader specialist deployments remain future plans or limited previews.
The question at launch is what a Dot can do while its owner is away.
A Dot Gets a Cloud Computer and Connected Apps
According to OpenAI’s Dots announcement, Dots run on GPT-6 Astra, have their own cloud computer and can access apps a user connects. The cloud computer gives an agent a place to work without requiring the user to keep a laptop open. Connected apps provide context and tools outside a ChatGPT conversation.
OpenAI describes a Dot as a persistent assistant that can keep pursuing a user-defined goal. Users start with a primary Dot they can name and personalize. Instead of starting each request from a blank prompt, they can assign ongoing work to an agent intended to remain available for it.
Persistence does not establish reliability on long-running tasks. TechCrunch’s account of the launch describes OpenAI’s examples of a developer monitoring customer feedback and a scientist investigating new experimental results. Those are proposed workflows, not independent tests showing that a Dot can safely ship a code change or reach a sound scientific conclusion on its own.
The examples also involve different levels of authority. Reading feedback or examining new data is one thing; editing a repository or changing a shared analysis is another. What an agent can do depends on the connected app, the permissions granted and OpenAI’s action rules as well as the model behind it. A cloud computer makes sustained work possible, but it does not grant unrestricted authority.
The First Dot Is Included, but Eligibility Is Limited
OpenAI says the first Dot is included for users on eligible ChatGPT Pro and Business Premium plans. Rollout is by market, so the feature may not have appeared in every eligible account. The announcement does not establish a separate price for that first included Dot. “Included” also does not mean unlimited use or free additional agents.
OpenAI specifies Business Premium, not every business subscription. Enterprise, Edu and Healthcare access is a beta that requires an administrator to enable it, rather than a general rollout to every member of those organizations.
A brief exchange in the OpenAI developer-community DevDay thread illustrates the uncertainty. One Pro subscriber reported that they could not find Dots, and another commenter pointed to geographic rollout restrictions as a possible explanation. That exchange is anecdotal. It says nothing about how many customers have access and does not confirm any particular account’s eligibility.
Prospective users need to check their plan, market and rollout status. A company also needs to know whether its workspace has beta access and an administrator has enabled it. Neither an eligible subscription nor an announcement-day launch guarantees immediate availability to everyone on that plan.
Background Research Has a Read-Only Boundary
OpenAI draws a specific line around one of Dots’ most consequential features. When a user is not actively working, the company says a Dot’s proactive research uses connected-app tools restricted to read-only access. The agent can gather or examine information through those tools without treating the owner’s absence as permission to make changes.
That rule applies to proactive research, not the whole product. OpenAI says other actions that affect an account are subject to rules and automated review. It does not follow that every possible edit, message or other account-affecting action always requires separate human approval. Automated review alone does not establish that an action is safe or that every mistake will be caught.
Consider a Dot researching customer reports while its owner is away. That fits OpenAI’s description of read-only background research. Making a change in response to those reports would depend on the requested action and the rules that apply. OpenAI’s example of an agent working through a developer’s feedback queue is not a promise of unattended deployment.
App connections deserve the same scrutiny. They can make an agent more useful while putting private material and consequential tools within reach. Users and administrators need to assess the permissions they grant for each task. OpenAI’s stated boundaries are safeguards, but people still need to know which accounts an agent can see and what actions it is allowed to attempt.
Texting and Teams of Dots Are Not Part of the Standard Launch
Some descriptions of Dots concern what OpenAI wants the product to become. The DevDay recap describes messaging and collaboration ambitions. The first broadly described offer, though, is one primary Dot for an eligible user. Text-message support is planned, and multiple Dots working together are a future direction, not features every customer can activate at launch.
Specialist Dots are in a separate, limited enterprise preview. The idea is to give agents defined organizational responsibilities, with appropriate identities, credentials and tools. OpenAI has also discussed planned integration with Microsoft Agent 365 governance controls. Neither the preview nor the planned integration amounts to generally available specialist-agent deployment.
TechCrunch reports that users can reach Dots through organizational platforms such as Slack and Microsoft Teams, while texting is still to come. Access through a messaging platform does not mean a user can create several specialist Dots there or give them unrestricted access to workplace systems.
OpenAI’s longer-term vision is understandable: an organization might eventually assign one agent to monitor an ongoing project and another to handle a different responsibility. Today’s buyer should evaluate the single Dot and its available controls, not an imagined team.
The Agent’s Job Matters More Than Its Avatar
Dots arrives amid a wider push to make agents feel like approachable assistants. TechCrunch compares its cartoon-like presentation with Meta’s Muse. The visual similarity says little about whether the products have equivalent access, safeguards or availability. For Dots, the consequential combination is persistence, a cloud computer and connections to services where real work happens.
Frequently Asked Questions
4 questions
1What Can an OpenAI Dot Do at Launch?
A Dot is a GPT-6 Astra-powered agent with its own cloud computer and access to apps its user connects. OpenAI is rolling out one primary Dot to eligible users. It describes ongoing research and other persistent work as intended uses, but its example workflows are not independent proof that every task can be completed reliably.
2
Sources
- OpenAI’s Dots announcementopenai.com
- TechCrunch’s account of the launchtechcrunch.com
- OpenAI developer-community DevDay threadcommunity.openai.com
- DevDay recapopenai.com
