Start Here

Dots, without the AI jargon.
Start with the everyday idea.

You don’t need to follow every AI announcement. Here’s what a Dot is, how it fits alongside other AI tools, and why you might want one.

Imagine a helper with a notebook.

You explain the job, share what matters, and agree on when to check in. The helper keeps working, brings you something to review, and asks when a decision needs you.

That’s the idea behind a Dot. The notebook is a way to picture its saved context. The helper is software: it can misunderstand you, miss details, or be wrong. You still check the work.

Our analogy, based on OpenAI’s introduction to Dots.

The AI words, in everyday language.

These words describe different parts of the picture. One product can be a chatbot, use an LLM, and run agents and automations.

LLM / model

The language engine. A large language model learns patterns from training data to work with language and generate responses. It can make mistakes.

An ingredient inside an AI product, rather than the whole product.

Chatbot

A way to talk with AI. You ask, it replies, and you steer the conversation. Some chat products also have memory, tools, and scheduled tasks.

“Help me think of a good menu for our neighborhood picnic.”

Agent

AI that can work toward a goal across several steps, using tools when available. How much it can do depends on the product and its permissions.

“Compare three picnic locations and prepare a shortlist.”

Automation

Work set up to run when something happens or at a chosen time. A simple automation follows fixed steps; more flexible ones can include an AI agent.

“Every Friday, remind me to check the guest list.”

Tools / connections

Ways an AI product can read information or do something beyond writing a reply. Each connection has its own access and permissions.

Access to a planning document, an app, or a browser.

Dot

OpenAI’s ongoing agent in ChatGPT. It can carry work between conversations and bring back results and questions that need your decision.

“Help keep the picnic plan up to date as we get ready.”

What makes Dots different?

A Dot is a kind of agent. Its emphasis is an ongoing relationship with your work. OpenAI describes a helper with saved context, its own cloud computer, connected tools, and the ability to continue between conversations.

You can give it new information or change direction as the job develops. It can also delegate parts of the work to other agents. See how tasks and memory work.

Other AI tools can also remember things, use tools, and do recurring work. The appeal of Dots is how those pieces come together in one continuing helper. It isn’t a promise that Dots is better at every task.

Why might I want one?

Because remembering to follow up can be a job of its own. You might want help gathering updates, keeping a plan current, or preparing the next draft while you focus on the people involved.

A useful fit: a recurring responsibility with information you can safely share and a result you can check.

A simpler tool may be enough: use a chat for a one-off explanation, a reminder for a deadline, or fixed automation for a predictable sequence of steps.

Try it when the help is worth the setup and review. You don’t need a Dot just to keep up with AI.

Ongoing help still needs clear limits.

Say what it may use, what it should produce, and which actions need you. Instructions and product permissions are different controls. Neither makes the output infallible. Read OpenAI’s guide to controlling your Dot.

The hope of Dots: what could be.

Two fictional stories about the kind of help we’d like to see. These are imagined scenarios, not customer testimonials or tested results. Actual results depend on access, permissions, instructions, and review.

Fictional story · A small business

Maya closes the shop with fewer loose ends.

Maya repairs bicycles. She’s good at the work, but notes about parts and customer questions pile up while she’s on the shop floor.

Imagine she gives her Dot permission to review one job sheet and asks for an afternoon list of unresolved questions. In this story, it spots a missing wheel size, groups the jobs waiting on parts, and prepares two reply drafts.

Maya checks the sheet, corrects a mistaken assumption, and sends the replies herself. Prices, promises, and orders stay with her.

The hope: less reconstructing the day, more attention for the customers in front of her.

Try the follow-up example brief

Fictional story · Everyday life

Jon gets to enjoy the picnic he’s organizing.

Jon volunteers to organize a neighborhood picnic. Every new reply changes the shopping list, and the backup plan is still scattered across his notes.

Imagine he shares a planning document and asks his Dot to review it twice a week. In this story, it prepares an updated supply list, notices that nobody has confirmed the tables, and drafts a rain-plan checklist for him to review.

Jon checks the counts and talks with the volunteers. The Dot doesn’t buy supplies, make reservations, or send invitations.

The hope: a plan that’s easier to pick up again, with fewer small details living only in Jon’s head.

Try the event-planning example brief

Product descriptions checked September 30, 2026 against the official guides linked above. The stories and analogies are our own. We have not tested these scenarios in Dots.

Can I use Dots yet?

Access is rolling out. Availability depends on your plan, region, and workspace settings. Check your account and OpenAI’s current setup guide before planning around a particular feature.

Check access and setup

What do we do?

Just the Dots helps you decide what to ask for, what context to share, and where the work should stop. Our free builder produces an instruction document.

About this independent guide

A simple starting path

Pick one thing worth preparing. Try a summary of open project tasks, a draft event checklist, or a comparison of three ideas. Choose something you can review yourself.
Provide a small, appropriate sample. Three fictional customer notes are enough to test a follow-up brief. You do not have to connect an entire account to decide whether the job makes sense.
Describe a good result. Name the format, what needs a source, and how missing information should be labeled.
Draw a clear line. Start with reading authorized information and preparing drafts. Require review before messages, purchases, deletions, access changes, or production changes.
Review, correct, and repeat. Check the first result carefully. Change one instruction and try the same small task again.

If you already have access

Follow OpenAI’s onboarding on the desktop app or desktop web. Mobile creation is not currently supported. Review available connections and their actual permissions before adding sources. Official setup guidance (checked September 30, 2026).

Keep your written instructions and product controls in sync. A brief does not grant access or configure settings. See the boundaries lesson before expanding the responsibility.

If you don’t have access yet

You can still define the job, gather an appropriate sample, and write down what a useful result looks like. The lessons and builder require no OpenAI account. Keep your brief to revisit when the product is available to you.

Start with a recipe, or start with your own idea.

Our six recipes give you a concrete starting point. They are example briefs, not installed agents or hands-on tested workflows. Read the boundaries, adapt the context, and try the small first test.

Choose a recipe Build my own brief