OpenAI Dots: Always-On Agents That Work Under Your Rules

Most artificial intelligence assistants wait for an instruction, deliver an answer, and stop. OpenAI Dots propose a different relationship: agents that can maintain context, work over extended periods, use authorized apps, and move projects forward while a person focuses elsewhere.

The idea is compelling, but it also requires careful interpretation. A Dot is not an infallible employee or a blanket authorization to act for its user. It is a system capable of taking on continuous work within permissions, rules, reviews, and boundaries that must be configured thoughtfully.

What are OpenAI Dots?

Dots are artificial intelligence agents designed to take on ongoing responsibilities. Rather than answering one isolated request, they can retain a project’s objective, notice changes, prepare work, and return with progress, questions, or decisions that require human involvement.

Each Dot has its own cloud computer and browser. It can also use plugins and apps connected by the user, within the permissions granted. OpenAI explains that Dots are powered by GPT‑6 Astra and can work across several projects without requiring the user to manage every step in separate conversations.

The word “persistent” does not mean unlimited autonomy. A Dot operates within built-in rules, app permissions, Custom Rules, and review systems. It may perform some actions independently, require approval for others, and hand the most sensitive steps back to the person.

A Dot does not replace the judgment of the person directing the work. Its value lies in preserving context, carrying out processes, and returning progress ready for evaluation.

The difference between a Dot and a regular ChatGPT conversation

In a typical conversation, the user presents a need and ChatGPT responds within the thread. A Dot adds operational continuity. It can follow a project, consult connected sources, coordinate tasks, and maintain a line of work even when the user is not actively chatting.

The central difference is not that a Dot “thinks on its own,” but that it has infrastructure for keeping work moving. It therefore resembles an ongoing collaboration more than a one-time query.

A cloud computer for continuous work

Each Dot operates inside a separate computing environment. It can browse, analyze information, create files, and use tools there. The user can open this environment to inspect its work and follow progress.

The cloud computer remains separate from the user’s personal device. If the user chooses to connect a computer, the Dot may access local files or apps within the permissions granted and the relevant safety checks.

This design lets the agent work without permanently occupying the person’s computer. It also makes it more important to define minimum access and review which information the Dot can consult.

How Dots learn preferences and standards

A Dot can learn how its user organizes work, which goals matter most, and what qualifies as an acceptable result. This adaptation develops through conversations, corrections, files, connected apps, and feedback on its output.

This can reduce repetitive instructions. A communications professional might establish tone, priority audiences, approved sources, verification requirements, and criteria for approving a statement. A creative professional could define brand identity, formats, restrictions, and review stages.

Learning preferences does not mean understanding them correctly every time. Standards should be documented, exceptions should be explicit, and important materials need review before publication or delivery.

Proactive research: helping before a request arrives

One of the most distinctive Dot capabilities is proactive research. When the user is not actively working with the agent, it can look for useful information in connected sources and prepare private notes.

OpenAI places an important restriction on this feature: proactive research tools are read-only. They cannot send messages, change app content, or control a browser or computer. If the Dot wants to act on a finding, it must follow the regular action rules and review process.

In practice, it might notice that a project requirement changed, new relevant information arrived, or a deadline is approaching. It can then present the finding and suggest an action, but the discovery does not automatically grant permission to carry it out.

Dots work across different channels

Dots can be used through ChatGPT on web, desktop, and mobile. OpenAI also supports interactions through Slack and Teams. The objective is to preserve context as work moves from an individual conversation into a shared team environment.

A Dot can send progress updates, request information, or ask for a decision. This reduces the need to check a dashboard continually to learn whether a task has finished.

Cross-channel continuity also creates responsibility. Before adding a Dot to a group space, an organization should define what information it may share, whom it may contact, and which actions require authorization.

Applications for communications, content, and creative work

Dots may be especially useful in processes where information changes and work must be updated. Possible applications include:

Organizational communication monitoring

A Dot could consult authorized sources, organize updates, prepare summaries, and flag matters requiring a response. It should not automatically publish an institutional position without clear rules and approval.

Content production

When an interview or recording arrives, the Dot can identify clips, prepare notes, suggest posts, and apply edits across several assets. The creator retains editorial control and approves the final materials.

Campaign updates

If a product, audience, or launch date changes, a Dot can locate affected materials, prepare updated versions, and flag contradictions. This helps prevent one document from being updated while the rest of the campaign retains outdated information.

Research and analysis

The agent can review new data, rerun analyses, and update charts or explanations. The person must validate the methodology and decide whether the evidence truly supports a change in conclusions.

Proposals and business development

A Dot can compare requirements, product documentation, and account history to prepare a proposal. Commercial commitments, prices, and terms should remain under the control of responsible people.

Permissions, Custom Rules, and approvals

Users decide which apps a Dot can access. Permissions are managed through ChatGPT connections and supplemented by built-in rules governing when the agent may act.

Custom Rules add further boundaries. For example, a user can instruct a Dot never to send email, to prepare social posts without publishing them, or to request approval before changing shared files.

These rules do not remove mandatory controls. Changing a password or transferring money are examples of actions that stay with the user. Other actions, such as permanently deleting data or installing software, may require confirmation every time.

What Auto-review does

Before certain actions take place, a separate system called Auto-review checks the proposed step against the user’s instructions, Custom Rules, and safety requirements.

If the action is permitted, the Dot can proceed. If the system detects a problem, it can block the step and explain why. The Dot may then ask for information, request approval, find an authorized alternative, or stop.

User approval cannot bypass every protection. Core requirements remain active, and the Dot cannot disable them from its own computer.

Memory and privacy: an important distinction

A Dot can receive memories and recent context from ChatGPT. Conversations with a Dot may also contribute to ChatGPT memory. Turning Memory off stops new sharing, but it does not automatically delete information that the Dot has already incorporated into its context.

Official documentation also states that users currently cannot inspect, modify, or delete each individual memory retained in a Dot’s context. Deleting that context requires deleting the Dot. Files and conversations created by the agent are stored separately and do not automatically disappear with it.

OpenAI does not use content from Business, Enterprise, and Edu workspaces to train its models by default. On personal plans, users can control this through the model improvement setting.

Risks and limitations of Dots

The ability to act increases the system’s value, but it also expands the consequences of a mistake. Key risks include:

• Misunderstanding the objective: a Dot may move forward with an incomplete understanding of the priority.
• Using outdated information: a connected source does not guarantee that all its data is correct.
• Applying a rule outside its context: a regular preference may not suit an exceptional situation.
• Sharing more than intended: an action may combine information from several authorized sources.
• Overdependence: delegating without retaining internal knowledge weakens the ability to supervise.
• Confusing activity with progress: completing many tasks does not mean the project is moving toward the right result.

OpenAI explicitly warns that Dots can make mistakes. Safeguards reduce risk, but they do not make every action correct or replace human review.

How to prepare a Dot for effective work

1. Define a specific responsibility: start with a bounded process rather than asking it to “handle everything.”
2. Describe the expected outcome: explain what completed, acceptable work looks like.
3. Connect only what is necessary: apply the principle of minimum access.
4. Create clear rules: distinguish preparing, editing, publishing, sending, and deleting.
5. Set approval points: identify which decisions need human review.
6. Test with reversible tasks: observe how it interprets instructions before allowing actions that are difficult to undo.
7. Correct with evidence: explain what worked, what failed, and how it should act next time.
8. Review permissions regularly: remove connections that are no longer necessary.

Personal Dots and specialist Dots for organizations

A primary Dot works on behalf of one person and learns that person’s goals and preferences. OpenAI has also previewed specialist Dots for organizations. These agents would have their own identity, credentials, and company-defined responsibilities.

A specialist Dot could take on a process such as support, invoice review, procurement, or email marketing. Implementation requires authorized systems, human owners, escalation rules, and audit mechanisms.

OpenAI has also announced work with Microsoft to integrate specialist agents with enterprise governance controls in Agent 365. This mode remains more limited than the personal experience.

Availability and access

OpenAI began rolling out Dots to selected plans and markets, while Enterprise organizations may access a beta experience when an administrator enables it. Availability, limits, and terms may change as the product evolves.

Users should therefore verify current information on the official page before subscribing to a plan or designing a process that depends on the feature.

Frequently asked questions about OpenAI Dots

Conclusion: delegating work also requires learning to govern it

Dots represent a shift from artificial intelligence that responds to artificial intelligence that maintains responsibility. Their cloud computer, connections, and ability to retain context can support projects that previously required constant supervision.

The benefit does not come from granting unlimited access. It comes from defining sound objectives, connecting only the sources required, setting rules, and reviewing decisions that may affect people, accounts, or information.

The most useful question is not how many tasks a Dot can perform. It is which work should be delegated, under what conditions, and with what evidence we can determine that it is actually being done well.

Official sources consulted

• OpenAI: Introducing Dots.
• OpenAI: How we build safety, security, and privacy into Dots.
• OpenAI Help Center: Dots privacy, security, and safety FAQs.
• OpenAI Help Center: Getting started with your Dot.