My First Day With OpenAI Dots: Easy Setup and a Very Likeable Assistant

I finally got access to OpenAI’s new Dots tool around 11 p.m. The setup was easy. Making my Dot cute took considerably longer, but that's just because it's easy to obsess over..

His name is Herman (the blue guy in the picture), and I got unnecessarily invested in his appearance.

Apparently, the quickest way to get me attached to an AI assistant is to give me creative control over its little face. Do with that information what you will.

Once I finished designing him, I connected my apps and started talking. By the next morning, I was calling him to continue the conversation.

That’s my strongest first impression of Dots: it feels natural to use. The setup is approachable, the conversation is responsive, and having a persistent assistant makes sense surprisingly quickly.

I’ve spent plenty of time working with agents. After my first night with Herman, I could see myself recommending this experience to someone who has never used one.

What is OpenAI Dots?

OpenAI introduced Dots in its September 29 DevDay announcements. A Dot is a persistent personal agent inside ChatGPT that can take on ongoing responsibilities, use connected apps, and keep working between conversations.

According to OpenAI’s documentation, Dots run on GPT-6 Astra and have their own cloud computer and browser. They can research, analyze data, prepare documents, and build software while you continue talking to them or go do something else.

The idea is to give you an assistant that carries context forward, follows ongoing work, and brings things to your attention.

We’ve already covered how ChatGPT Work makes agent capabilities more accessible. Dots gives those capabilities a persistent presence you can message or call throughout the day.

My main question going in was whether that would feel meaningfully different from opening ChatGPT.

For me, the answer is yes.

The best setup step was letting Herman interview me

Once my apps were connected, I let Herman ask me questions.

We discussed what I’m working on, my priorities, how I like to work, and what I want help with.

I’ve used this approach with agents for a while. Rather than trying to write one giant prompt containing every preference I’ve ever had, I let the agent interview me. It’s easier to answer a useful question than anticipate everything it might need to know.

That worked particularly well here because Herman is supposed to stick around.

OpenAI says Dots use relevant ChatGPT memory and keep their own notes about preferences, decisions, and ongoing responsibilities. Spending a few minutes giving Herman context felt like a sensible investment in something I’ll keep using.

The conversation itself was quick and pleasant. He seemed to understand what I meant and what I was trying to accomplish without making every exchange feel like a prompt-writing exercise.

The interface helps, too. Blue bubbles and a simple thread give it an iMessage feel. It’s familiar.

I found myself saying please and thank you and asking for things conversationally. I do that with other agents, too. I don’t see much reason to be rude just because the other side is software.

Having an assistant that notices is the interesting part

The biggest difference for me was the combination of persistence, proactivity, and context.

Take something ordinary, like an email about a Ray-Ban warranty claim. Herman can notice that kind of item and ask whether I want help responding or getting a repair arranged.

That’s useful because so much personal admin gets stuck between “I should deal with that” and actually dealing with it.

I can ask ChatGPT for help with a warranty claim. Having an assistant notice the loose end and offer to help changes how that work enters my day.

OpenAI documents proactive research across information a Dot has permission to read. That research can lead to a suggestion or question; follow-up actions use the applicable permissions.

This is the part I’m most interested in living with over time. How much useful work can Herman bring to my attention that I wouldn’t otherwise remember to ask about?

Even on day one, the ongoing-assistant format felt compelling.

A small lesson in telling your agent how you work

I also got an early reminder to be clear about my preferences.

I asked Herman to deal with a simple email. He wrote a short, perfectly acceptable reply that sounded like me—and sent it before I had a chance to review it.

I had a brief “oh no” moment, read the message, and realized it was completely fine.

Then I told him that, going forward, I want to see outgoing messages before they’re sent.

That was the lesson: establish the workflow you want. Some people may be happy to let an assistant handle basic replies immediately. I prefer to review them for now, and I can imagine relaxing that preference as we work together.

OpenAI provides instructions and optional custom rules for setting those boundaries. I’m cautiously optimistic that my preference will stick, based on how similar instructions have worked with other agents I use.

The experience didn’t change my enthusiasm for Dots. It gave me one more thing to explain to Herman.

I gave him a usage policy, too

Another early conversation was about getting the most useful work from the subscriptions I already pay for.

OpenAI says Dot conversations don’t count toward ChatGPT usage limits, while tasks a Dot starts or manages in Work or Codex count toward those products’ limits. Deeper work has a plan allowance, with extended limits during the first month after launch.

So I gave Herman a rough policy: use ordinary conversation when it’s sufficient, and bring in stronger tools when the job needs them or the simpler route produces poor results.

I enjoyed turning that fuzzy preference into an ongoing instruction. It’s the kind of configuration I’d much rather explain in a conversation than build into a workflow by hand.

It will take weeks to know how consistently he follows it or whether it makes a measurable difference to my usage. For now, setting it up was easy.

Calling Herman made me want one more feature

I called Herman the next morning.

The voice appeared to use my regular ChatGPT voice selection, but the context and behavior gave the conversation a somewhat different feel.

I’d love to assign my Dot a separate voice.

If I’m going to speak regularly to both ChatGPT and Herman, having distinct voices would help give each experience its own identity. It’s a small product detail, but one I noticed quickly.

More broadly, the fact that I went from setting him up at night to calling him in the morning says something about how comfortable the experience felt.

Who I’d recommend Dots to

I’m especially interested in Dots as a first agent for people who haven’t tried one.

Traditional agent tooling can sound intimidating. Start talking about APIs, MCPs, workflow builders, policies, plugins, and skills, and plenty of people will reasonably decide they have other things to do.

Dots makes the starting point much easier: connect your apps, have a conversation, and give your assistant something to help with.

There’s something here for experienced users, too. I like the natural-language setup, the ability to explain standing preferences, and the possibility of using Herman to coordinate tools I already use.

But the broader opportunity is getting more people comfortable with ongoing AI assistance. This is an experience I can imagine explaining to a friend without turning the explanation into technical support.

I’d start with calendar review, email triage, research, drafting, and everyday admin. Give it context, explain how you want it to work, and build from there.

My day-one verdict: I really like it

Dots is responsive, fun to use, and more natural than I expected. The form factor makes sense to me.

There are questions I’ll revisit after a few weeks: how well Herman follows ongoing preferences, how useful his proactive suggestions become, and what sustained usage looks like. I can also imagine wanting specialized Dots for different responsibilities, though I don’t yet know whether several would be better than one assistant with good context.

Those are things I’m curious to explore as I keep using it.

After day one, I’m happy to recommend trying Dots if you have access—especially if you’ve been curious about personal agents but put off by the setup.

The question I keep coming back to is whether putting these capabilities into a persistent, conversational assistant makes me use them more often and more naturally.

So far, Herman is making a pretty good case.

Getting started is refreshingly easy

The functional setup felt like roughly five minutes. Most of it was questions, clicks, and permissions.

I connected Google Calendar, Google Docs, Google Drive, and other available integrations. Then I started telling Herman about my work.

If you can set up a ChatGPT account, you can probably set up a Dot.

Here’s the practical starting point, following OpenAI’s setup instructions:

1. Open ChatGPT on desktop web or in the desktop app. When Dots appears in your account, open it and follow the introduction.
2. Connect the apps you want it to use. You can add connections later, too. Connecting your computer is a separate option in the desktop app.
3. Give it a name and personalize its appearance. It starts with a default you can change after setup.
4. Talk through your priorities and preferences. Explain what you do, where you want help, and which actions you want to approve.
5. Give it something useful to handle. Start with a calendar review, some research, a draft, or a small administrative task.

On desktop, your Dot’s profile includes Activity for reviewing delegated work and Scheduled for recurring tasks. OpenAI’s controls guide explains where to inspect progress and manage permissions.

As of September 30, Dots is rolling out gradually to eligible Pro 100, Pro 200, and Pro 500 users over 18 outside the EEA, UK, and Switzerland, plus Business Premium and Enterprise worldwide. Enterprise administrators must enable it. You create your Dot on desktop; mobile-app interaction depends on the supporting update, and mobile web isn’t supported. The availability page has the current requirements.

My access arrived late that evening, so being eligible doesn’t necessarily mean seeing it immediately.

Related reading

• GPT-5.6 and ChatGPT Work: Everything OpenAI Announced — Background on the tools Dots can coordinate and OpenAI’s push to make agents more accessible.
• Watch Us Test ChatGPT Agent Live for the First Time — Our earlier hands-on experience offers a useful comparison with Dots’ persistent assistant format.