OpenAI Dots: What We Know—and What Remains Unconfirmed—About Its Agents

A form of assistance that goes beyond a one-off answer

OpenAI describes Dots as assistants that can proactively help with complex projects and everyday tasks, and says they keep work moving forward. This general description is the most solid information in the sources provided: it comes from a company page titled “Introducing dots.” However, the available summary of that page does not explain how tasks are carried out, which systems the agents can use, or which decisions remain with the user.

The difference between an assistant that answers a question and an agent that continues a task in the background matters. In the latter case, it is important to understand not only what goal the agent is given, but also which tools it can use, for how long, with what permissions, and how its actions can be corrected or stopped. A general promise of continuity, by itself, does not answer these practical questions.

Several news reports describe Dots as persistent agents that can work while the user is offline or have their own cloud computer. Those descriptions help identify the approach attributed to the product, but they are not equivalent to official technical documentation. Based on the material available, it is not possible to establish that each of these features is confirmed, or to define exactly what it means in practice for work to continue “in the background.”

Persistence, a cloud computer, and interaction: claims that need context

The available coverage presents three related but distinct ideas: that the agents are persistent, that they have a cloud computing environment, and that they can keep working without direct supervision. These ideas should not be treated as interchangeable. A service may retain the context of a project without continuously carrying out tasks; it may also offer a remote execution environment without giving an agent permission to act across every connected service.

The descriptions from El Mundo and The Decoder characterize Dots as agents that would operate continuously and have cloud computers. TechCrunch, meanwhile, describes them as agents that would pursue user-defined goals in the background with minimal supervision. These are journalistic claims about the product, not a complete specification of its capabilities. The sources provided do not allow us to verify the conditions, limits, or exceptions that would apply to this kind of operation.

The Decoder’s description adds that the agents could be accessed through ChatGPT, Slack, and Microsoft Teams, and that they would use read-only access when nobody was interacting with them. That is relevant detail, but it is not supported by the available summary of OpenAI’s official page. It is therefore not possible to present those integrations or that permission mode as features OpenAI has confirmed on the basis of this material.

Nor can we infer that an agent with read access poses no risks, or that a permission granted for one task would apply to every other task. To assess the actual level of control, we would need to know the scope of each permission, whether it can be limited by application or action, which operations require approval, and what records are available for review.

The central question: what can it do without approval?

For anyone considering a persistent agent, the decisive question is not simply whether it can work autonomously, but which actions fall within that autonomy. Reading information, preparing a draft, editing a file, sending a message, and completing an external transaction all have different levels of impact. The sources provided do not specify which category Dots’ actions fall into or which ones require confirmation.

The material also does not describe specific controls for pausing a task, revoking access, reviewing an action history, or changing priorities while the agent remains active. A report claiming that users can stop, approve, or inspect any operation would need direct support in product documentation. That support is not included in the material available here, so these controls should be treated as unverified, not as confirmed capabilities.

The same caution applies to the read-only access cited by The Decoder. If confirmed, it would be an important detail for understanding how the system operates when there is no active interaction. Even then, we would still need to know which data it applies to, whether it is the default mode, how it changes when an action is authorized, and what options users have to restrict it. The label “read-only” does not replace that information.

Announcement and availability: not enough information to establish the scope

The news reports provided place the presentation of Dots at DevDay 2026, and several are dated September 29, 2026. The official page supplied is titled “Introducing dots,” but the available summary does not include a publication date or independently confirm that event context. To be precise, the date and presentation setting should be attributed to those reports rather than treated here as independently verified by the primary source.

The material also does not establish whether the product is available to all users, only to businesses, on certain plans, or in specific regions. A secondary source in Korean characterizes it as an enterprise-focused offering, while other descriptions present it as a personal assistant or a feature integrated into ChatGPT. The available summary of official information does not resolve this difference in how the product is positioned.

These sources do not confirm a definitive list of integrations, a rollout schedule, eligible countries, or access requirements. Answering those questions would require a complete product page, help documentation, or an official announcement that specifies the terms. Until then, examples in news coverage should not be confused with features already available to the public.

What could change for users—and what is worth waiting to learn

If Dots can keep a task moving between sessions, the potential change would be continuity of work, not merely the quality of an answer. In principle, an assistant could retain the context of a project and make progress on delegated tasks. But the sources provided are not enough to describe that experience precisely or to guarantee that the system can act independently in external applications.

Comparisons with other persistent agents also require caution. Some reports mention Meta Muse as a point of comparison or competitor, but the material available does not provide a verifiable comparison of features, permissions, performance, or availability. A shared idea of agents working in the background does not demonstrate that the products have the same design or controls.

For now, the clearest conclusion is a limited one: OpenAI presents Dots as proactive assistants for complex projects and everyday tasks, while various publications attribute persistence, cloud execution, and application integrations to them. The first point is supported by the summary of the official source; the operational details come from coverage that is not confirmed by the primary documentation included here. Permissions, approvals, available tools, and rollout remain open questions.

For anyone deciding whether to delegate work, the practical checks are straightforward: look for current documentation on permissions by application; distinguish reading from writing or sending; check whether an action requires approval; confirm how to stop a task; review the action history; and verify availability for the relevant plan or region. Without those answers, it is more responsible to treat Dots as a proposal for proactive assistance whose specific operation cannot yet be established from these sources.

Keep exploring

Corrections and transparency

If you spot incorrect or outdated information, send us a correction with the page and source we should review.