OpenAI Dots: What They Are and How to Use Them

You have a course draft in one place.

A client's corrections in another.

And a list of things to update after the next workshop.

The hard part is keeping the work moving when you are doing something else.

That is the kind of problem OpenAI is aiming at with Dots.

Here is what they are, where to start, and a few jobs worth testing.

Product details checked October 1, 2026. The examples below are proposed workflows, not results from testing Dots.

What are OpenAI Dots?

OpenAI introduced Dots on September 29, 2026. They are ongoing agents in ChatGPT, powered by GPT-6 Astra, with their own cloud computer and access to apps you connect.

You give a dot a goal and boundaries. It can continue working between conversations and bring back work for review.

OpenAI's examples include preparing content drafts and updating project materials as requirements change. Read the announcement.

For an instructional designer, the interesting question is what responsibility to hand over.

Not another request to write a course.

Something specific that keeps falling between the other things you have to do.

What could an instructional designer try?

Start with approved sample material. Leave confidential company and learner information out of an experiment unless your organization has approved the environment and access.

Check whether the lesson still matches the objective

Imagine a lesson on handling a frustrated customer. The objective says the learner must decide when to escalate.

The draft has a clear explanation of escalation. But the practice questions only ask learners to recognize definitions.

A useful assignment would be to compare the objective, lesson and questions. Ask for a table showing where the learner actually makes the required decision.

Require an exact reference for each finding. Which question? Which paragraph? What is missing?

You and the subject-matter expert still decide whether the practice is good enough. The proposed role is to find gaps you can inspect.

Track changes across a course package

Suppose an approved procedure changes. The same instruction appears in a lesson, job aid and facilitator notes.

Give the dot the old and new approved versions, plus copies of those three documents. Ask it to list each affected passage and draft the changes.

Include the version date with every finding. If two sources disagree, ask it to flag the conflict rather than choose a policy.

Check that it found the places you already know about. Then check the additional ones.

That gives you a way to assess its work before trusting a larger update.

Prepare the next step after a workshop

With an approved transcript and resource list, a proposed assignment could produce a draft follow-up note, a short FAQ and a list of unanswered questions.

Make it separate what the presenter actually said from suggested additions.

A helpful FAQ answer should point back to the source. A missing answer should stay missing until someone supplies it.

You review the content and decide what goes to attendees.

What could a small business hand over?

The same approach works for the small tasks around the work.

For example, provide a sample project tracker and ask for a list of overdue deliverables, conflicting dates and items waiting on a decision.

Ask for evidence beside each item, and keep the source tracker unchanged during the first test.

Or provide a proposal and updated client requirements. Ask for a comparison: what still matches, what needs clarification, and what might change the scope.

Keep pricing and commitments with you.

Another useful experiment is turning one approved article into several draft assets. An email introduction, a LinkedIn post and a workshop outline should carry the same facts, dates and links.

Ask the dot to list those shared details so you can check them once, then inspect each draft.

These are jobs to test. They are not promises that a dot understands your business or can use every system you own.

How do you get started?

Dots are rolling out gradually. As of October 1, eligible plans are Pro outside the EEA, Switzerland and UK; Business Premium in supported ChatGPT regions; and Enterprise beta with admin enablement, including Edu and Healthcare.

Create your dot in ChatGPT on desktop web or in the desktop app, including Windows. Mobile-app access follows setup where available; mobile creation and mobile web are not supported.

The first dot is included with Pro or Business Premium. Deeper work has an allowance, with extended first-month limits. It is not unlimited. Check current access and setup details.

Choose one responsibility before connecting more tools. Give it the smallest set of approved sources needed for that job.

Local-computer access is optional and starts off. You do not need to grant it for your first document-based experiment. The same setup guide explains how to enable or revoke it.

For an Enterprise account, ask the workspace owner to check the controls. Cloud access, connected-app permissions and local access are separate. See the workspace settings.

What should the first prompt say?

Try this with fictional material or an approved practice copy:

Read the report against your source material. Check a known gap, a section that is already correct, and an ambiguous case.

Did it find the gap? Did it invent another? Can you follow its references?

Then decide whether to give it the next draft. Add a recurring schedule only when the responsibility and review process are clear.

What should stay under your control?

OpenAI distinguishes proactive research from authorized actions. Proactive research can read permitted sources and save notes. Its tools cannot send messages, change app content or control a browser or computer.

Custom Rules cannot override core safeguards or Autoreview. Dots can still make mistakes. Read the permissions and safety FAQ.

For client work, define who approves changes and what approval covers. A draft you like is not permission to send the next ten messages.

Check data handling too. Disconnecting an app does not erase information already retained in the dot's context. Business, Enterprise and Edu data is not used for training by default; personal-plan settings govern model improvement. Review memory and privacy controls.

OpenAI's specialist dots are currently enterprise pilots. Teams of dots and additional capacity are future plans, not features to assume you have today. See what is current and planned.

Pick one recurring piece of work. Define the sources, the output and the decisions that stay with you.

Then check whether handing it over leaves you with less work after review.