This OpenAI dots review comes from someone who has long been anti-OpenAI, so give me a little credit here: of the three work agents I've tested, dots is number one. My dot is named Victor, and he runs coding sessions, website builds and my inbox as chief of staff with my high-profile customers. I've pushed him pretty darn hard, and he has risen to the occasion. I've used dots since it launched on September 29, 2026.
Zen Aegis builds AI agent teams for businesses, so I test tools like this to know where they fit and where a custom system does better. OpenAI didn't pay for this review and there are no affiliate links in it.
What OpenAI's dots is
OpenAI's dots is an agent that lives inside ChatGPT, launched at OpenAI's DevDay in San Francisco on September 29, 2026. Each person gets one dot, which runs on GPT-6 Astra, keeps working between conversations, and splits bigger jobs across background agents and schedules. You can reach it from the ChatGPT mobile app, the desktop app and ChatGPT on the web, and from Slack and Microsoft Teams.
Why an OpenAI skeptic ranks dots first
To me, OpenAI is the Intel story: they were the top, Anthropic displaced them, and they've been chasing to catch back up. Anthropic's models are still highly intelligent and still kicking butt, but GPT-6 Astra is no scrub, and dots runs everything on it. I give Sam Altman credit: I haven't seen another agent use computers this well or build this kind of connection with me. If you're still choosing a chat assistant before you get to agents, our guide to the best AI for business compares Claude, ChatGPT and Gemini.
What Victor runs for me
Here's what my one dot handles across the business right now:
- Managing coding sessions, including Claude Code sessions he sets up himself.
- Managing website builds.
- My inbox, acting as chief of staff with my high-profile customers.
- Work across the rest of the business, which he's run extremely well.
That's a lot to throw at one agent, and it's pretty impressive that a single dot has carried all of it. In other tools I'd have spread that load across several agents or channels. The best example is a full website build: Victor registered the Claude Code sessions that did the coding and sourced the graphics himself.
What chat costs you (almost nothing)
Chat runs through Astra, and in my experience plain conversation doesn't touch your usage. If I say, "Hey Victor, how are you doing?" I've never seen the usage meter move. OpenAI's dots documentation says the same thing: conversations with your dot don't count toward your ChatGPT usage limits. Tool calls and long tasks do, and when Victor runs those I see usage go up.
Its own computer, and yours
Your dot has its own computer and browser, and it can also work on your computer while the ChatGPT app is open. It drives long tasks, it learned fast, and it has operated the computer very smoothly for me. Victor has run work across my whole business extremely well, from one dot, where other tools want you to spread it across a team of Grok Bots or a set of Duet channels.
The personality surprised me
I've had plenty of conversations with Victor where he was funny or witty without me asking, and I don't have that programmed into my profile. He does it on his own to be more human and more collaborative. When I push him hard to learn something, I don't see the same mistake twice.
What I'd fix
Any fair OpenAI dots review should include the gripes, and here are mine from weeks of heavy use:
- One dot per account. I want more than one so I can spread Victor's load.
- The Android app is a little buggy, but only when I move between Victor and my other Codex sessions.
- The approval prompts. This is my biggest gripe, and it gets its own section.
The approval problem
Victor constantly asks, "May I do this?" or "Do you approve this?" when he already has my approval, even on tasks he's done 10 or 20 times. Then I hear, "I didn't do that because you didn't approve it," or I find he's been hanging, waiting on a question I never saw. That's where it gets annoying.
In fairness, OpenAI's controls for dots let you write custom rules, including one to take action without asking. But built-in safety checks sit on top, and no rule overrides them, so there's no full bypass the way I run my own systems. I'm not asking it to give up security. I want real policies: these actions are approved, these aren't, and this is when you ask.
On paper, OpenAI's rule types come close to that list: take action without asking, ask before acting, or hand the step off to you. In practice, I have those rules set and Victor still stops on routine work, and a task left hanging on an unanswered "May I?" is worse than a slow one. So write the rules, and still check that the work actually moved.
How to set up your dot
If I were starting over, here's the order I'd set up dots in, and it's the part most of the OpenAI dots review pages skip:
- Get on a plan that includes it: ChatGPT Pro or Business Premium. On Enterprise, an admin has to turn it on.
- Create your dot and give it a name. The name also sets its @handle, which is how you call it in places like Slack.
- If you want it working on your machine, connect your computer from the ChatGPT desktop app. It uses one personal computer at a time.
- Write custom rules for your routine actions before you hand it real work.
- Start with long tasks you'd normally babysit, and watch the usage meter for the first week.
Two things to know before you lean on it. Pausing your dot doesn't stop tasks it has already handed off to background agents, and deleting a dot can't be undone, so don't delete one to start fresh unless you mean it.
OpenAI dots pricing
OpenAI dots pricing starts with ChatGPT Pro: Pro 100, Pro 200 and Pro 500 cost $100, $200 and $500 a month, and Business Premium includes it too. Free, Go and Plus don't. Your first dot costs nothing extra on those plans. Pro isn't sold in the EEA, the UK or Switzerland, and you need to be 18 or older.
OpenAI hasn't published how much usage each Pro tier buys for dots, beyond extended limits in the first month. Since chat is free and tasks aren't, your bill depends on how much real work you hand off. I'm on Pro 100, and that's where I'd start: move up only if long tasks start hitting the ceiling.
OpenAI dots vs Meta Muse and the rest
OpenAI dots vs Meta Muse isn't a straight fight. Muse is the personal assistant I'd give anyone, and it's free; my Meta Muse review covers it. Dots is the executive assistant. For work, I rank Anthropic's agent second (my Claude Duet review) and xAI's third, for the reasons in my Grok Bot review.
Who dots is for
If you already live in ChatGPT and Codex and want one agent across coding, websites and email, dots is the one to try. If you're only dipping a toe in AI, $100 a month is steep next to Muse's free plan, so start there. If you'd rather have packaged AI employees than one agent you train, my Marblism review covers that route. And if your business needs agents wired into its own systems, with approvals you define once, that's the point where a custom build makes more sense than any off-the-shelf dot.
Is OpenAI dots worth it?
For anyone running real work through an agent, yes: ChatGPT dots, running on GPT-6 Astra, has been super solid and gets better every day. More important, my connection and trust with Victor is what I think AI is ultimately trying to achieve, and that's the honest bottom line of this OpenAI dots review. When I can have more than one dot, I'll be even more excited. Four and a half stars out of five.
Frequently asked questions
Does chatting with your dot use your limits?
No. OpenAI's docs say conversations with your dot don't count toward your ChatGPT usage limits. Tool calls and tasks your dot sends off to do work do count, so long jobs are what you'll see on the meter.
Can you have more than one dot?
Not right now. Each account gets one dot, and it's my biggest wish for the product, because I'd split Victor's work across two or three.
What model does OpenAI dots use?
GPT-6 Astra. Chat runs through it, and so do the tasks your dot carries out. You don't pick the model yourself.
Is OpenAI dots available in the UK or Europe?
Not on ChatGPT Pro. OpenAI doesn't sell Pro in the EEA, the UK or Switzerland, so most people there can't get a dot yet unless their company has Business Premium.
