OpenAI Dots for Work: What to Delegate, What to Review, and How to Get Started

OpenAI introduced dots on September 29, 2026, as always-on agents that can continue working between conversations. A dot has its own cloud computer, can use apps you choose to connect, and can bring work back for your review. That sounds different from opening a chat, asking one question, and starting over next time. But a useful first question is more specific: What responsibility would you give a dot, and what would you still want to approve yourself?

Here is what OpenAI says dots can do, who can access them at launch, and how to frame a real work assignment. We’ll use a product-launch example to make the boundary between ongoing monitoring and a finished decision brief concrete.

What is an OpenAI dot?

A dot is a personal agent in ChatGPT, powered by GPT-6 Astra. According to OpenAI’s launch announcement, it has a cloud computer and browser, can work with connected apps, and can carry context across tasks. You can message or call it in ChatGPT on desktop and web, and in the mobile app when access is available for your account after setup; OpenAI also describes messaging through Slack and Microsoft Teams.

The distinction is continuity. OpenAI describes a dot that can work on several projects, update you when it needs a decision, and learn from your feedback. In one of its launch examples, a product-launch dot revises draft materials as scope changes. In another, it looks at new evidence and updates an analysis for human review. Those are OpenAI’s examples of intended use, not a promise that every account has every integration or that a dot’s output is automatically correct.

A dot’s personal role is also different from the specialist dots OpenAI previewed for organizations. Specialist dots with dedicated responsibilities are starting in focused enterprise pilots; they are not the default self-service dot you create today.

Who can use dots, and where do you create one?

OpenAI says it is rolling out dots to Pro and Business Premium users, while Enterprise workspaces can try the beta when an admin enables it. The feature may not have reached every eligible account yet. The first dot is included with a Pro or Business Premium plan at no extra charge; deeper work still has a plan allowance.

To create the first dot, use the ChatGPT desktop app or desktop browser, follow the setup prompts, name it, and review any app connections and their permissions. After initial setup, OpenAI says you can message it in the ChatGPT mobile app when mobile access is available; you cannot currently create one on mobile, and dots are not supported on mobile web. Its getting-started guide explains how to inspect activity, review scheduled work, add custom rules, and pause the dot. If you do not see the option, check your account’s eligibility and rollout status before assuming your device is the problem.

What can a dot do without asking you?

That depends on the connected app, its permissions, the action, your rules, and any workspace policy. OpenAI says you choose which apps to connect. It describes built-in rules and Custom Rules that can allow a supported action, require approval, or block it. You can inspect progress and redirect the work. Your own computer stays separate from the dot’s cloud computer unless you explicitly connect it.

There is an important difference between a dot working in the background and taking an outward action. OpenAI calls the former proactive research: it can use read-only tools in apps you have connected to look for useful information, but that mode cannot directly message someone or change app content. Other tasks follow the app permissions and review rules you set. Decide which changes you want to check before they happen, and verify consequential results. OpenAI describes these controls in its Help Center.

For a product manager, that suggests a practical division: let the dot surface a change in customer feedback or a newly available document, but review the evidence, proposed wording, recipient, and commitments before anything consequential is sent or published. OpenAI itself cautions that dots can make mistakes. An approval prompt is a control, not a substitute for checking the work.

A work example: track a launch, then prepare a decision

Suppose your team is preparing a release. Feedback is coming through an approved support source, the product spec is changing, and leadership wants to know whether to adjust the launch message. OpenAI’s own launch example describes a dot learning a team’s audience and standards, then revising launch materials when scope changes. To turn that broad scenario into an assignment someone can supervise, define five things up front:

1. Goal: Flag changes that may alter the launch claim or timing; prepare a proposed update rather than publish it.
2. Sources: Name the specific connected feedback and product sources the dot is allowed to use. Do not assume access to every app simply because OpenAI’s plugin ecosystem supports many apps.
3. Trigger: Explain what counts as a material change—for example, a confirmed feature removal or repeated customer confusion about a promised capability.
4. Review boundary: Ask for source links, the proposed change, unresolved questions, and your approval before outward-facing messages or edits.
5. Done criterion: A short change brief that says what changed, who is affected, what needs a decision, and what remains unverified.

A starting instruction might read:

Watch the launch information I’ve authorized you to access. When a confirmed scope change or repeated customer issue affects our release claims, prepare a brief with the original claim, the new evidence, affected materials, and a recommended revision. Cite the source of each important claim. Do not send a customer message or publish a change without my review.

Treat this as an illustrative assignment, not a tested dot recipe or a guarantee that a particular app supports every step. Check the available connection and permission controls in your own account. As the work comes back, verify whether the feedback is representative, whether the spec change is final, and who owns the launch decision. Those are judgment calls a polished draft cannot settle on its own.

Make the finding useful to the launch team

The change brief is a starting point, not the launch decision. A product manager still needs to check the approved spec, see whether the customer feedback reflects a wider issue, and identify which claims and materials need review. The useful question is: what can the team say with confidence now, and what is still unresolved?

If you need to turn those materials into a reviewable recommendation, you can make that a separate, scoped task in MuseWork. Give it the approved spec and feedback you are permitted to share, along with any relevant public release information, and name the people who will review the result. MuseWork Deep Research can work from your files and public sources to organize evidence, flag gaps, and produce a structured report. A request could be:

Using the approved spec, customer feedback, and public release information I provide, check which launch claims still have support. Show the source for each finding, separate confirmed changes from open questions, and recommend which materials the product and marketing leads should review before publication.

The team can then check the report against the sources and decide what to change. If that is the result you need, start a research task in MuseWork with the materials you choose to provide. It is one way to move the launch work forward, not a required step in using dots.

Before you delegate, decide what “done” means

The strongest first assignment for a dot is not “handle the launch.” Give it approved sources, a trigger, a review point, and a concrete output. When the work becomes a decision, ask for the evidence, open questions, and a proposed next step in a form a person can actually review. That is how continuous attention becomes work the team can move forward.

References

1. OpenAI: Introducing dots
2. OpenAI Help Center: Getting started with your dot
3. OpenAI Help Center: Dots privacy, security, and safety FAQs