From answering to owning the job
OpenAI Dots are always-on AI agents inside ChatGPT that take on ongoing responsibilities, not just one-off prompts. OpenAI launched them on 29 September 2026 at DevDay, three weeks after Meta launched Muse.
A few weeks back I wrote about Meta Muse and what it means for brands. Muse showed us AI that can go out and do things for you. Dots take the next step. You don’t just give a Dot a task. You give it a job, and it keeps doing that job while you are busy with something else.
Think about the last three years like this:
- AI that answers. You ask ChatGPT a question, it replies, the chat ends.
- AI that acts. You ask an agent to book, buy or fill a form, it does it, the task ends.
- AI that owns a responsibility. You tell it what to keep an eye on, and it keeps working, week after week, and comes back when it needs you.
Dots are OpenAI’s big bet on stage three. In this post I’ll explain what Dots are, how they work under the hood, how they compare to Muse, and the part most people are missing: what happens to your website when agents start doing the browsing and buying.
What is OpenAI Dots?
A Dot is a personal AI agent that lives inside ChatGPT, runs on its own cloud computer, and works toward your goals around the clock. OpenAI calls them always-on agents built to handle everything (OpenAI).
Here are the basic facts from the launch:
- Launched: 29 September 2026, at OpenAI’s DevDay in San Francisco (TechCrunch).
- Model: GPT-6 Astra, the new model OpenAI released earlier in September (CBS News).
- Who gets it first: ChatGPT Pro and Business Premium users in eligible markets, with Enterprise, Edu and Healthcare able to try a beta once their admin turns it on.
- Cost: your first Dot comes with the plan at no extra cost. OpenAI says you’ll be able to add more Dots later.
- Where you talk to it: ChatGPT on desktop, web and mobile, plus Slack and Microsoft Teams. You can message it or hop on a voice call, and texting is coming soon.
- What it can reach: more than 4,000 apps through ChatGPT’s plugins.
One small detail tells you a lot. You create your Dot, give it a name, and it introduces itself. OpenAI is clearly not selling this as a feature. It’s selling it as a colleague. Sam Altman compared it to handing work to a high-agency engineer or a chief of staff (CBS News).
OpenAI also previewed specialist Dots for companies. These get their own identity, credentials and access to a company’s systems, so they can own a defined job like invoice processing or customer support. OpenAI is starting these as enterprise pilots and working with Microsoft to manage them through Agent 365.
Task vs responsibility: the one idea to understand
The real shift with Dots is simple: you stop giving AI tasks and start giving it responsibilities.
“Analyse my competitors” is a task. You ask, you get an answer, it’s done.
“Keep track of my competitors and tell me when something important changes” is a responsibility. It never really ends.
This is why OpenAI talks about Dots bringing you work “sometimes before you even think to ask” (OpenAI). When you are not working with it, a Dot looks for ways to help using the apps you’ve connected. OpenAI calls this proactive research.
One example from the launch says it well. An early tester’s Dot noticed he had forgotten to invoice a publication. It prepared the invoice and sent it once he approved. Nobody asked it to check invoices. It just knew that was part of looking after his work.
So the old question was, “How do I ask AI the right question?” Then it became, “How do I get AI to do this task?” Now the question is, “What can I hand over to AI for good?”
How Dots work under the hood?
A Dot is six layers stacked together: where you talk to it, a brain, memory, a safety check, its own computer, and the apps it acts in. The animation at the top of this post follows one request down through these layers and back up to you.
- Where you talk to it. ChatGPT, Slack or Teams, by message or voice. The Dot carries the same context across every channel, so you can start something in ChatGPT and continue it in Slack without repeating yourself.
- The brain. GPT-6 Astra does the thinking: planning, deciding the next step, and judging when a result is good enough.
- Memory. The Dot learns from your feedback over time: your preferences, how you think, and what good work looks like to you. This is what lets it do the work “your way” instead of a generic way.
- Safety check. Before any action that could affect your accounts or share information, the Dot runs an auto-review. It checks the action against your instructions, your Custom Rules and OpenAI’s safety requirements, then decides: go ahead, ask you first, or leave it to you.
- Its own computer. Every Dot gets its own cloud computer and browser. Your laptop stays separate unless you choose to connect it. You can open the Dot’s computer anytime and watch what it’s doing.
- The apps it acts in. Through ChatGPT plugins it can reach more than 4,000 apps. Plus, with its own browser, it can go to any website just like a person would.
Then it comes back to you. A Dot can message you with progress, questions, or a decision it needs from you. You can also follow all its background work in what OpenAI calls the Activity View.
The important thing to notice: layers 5 and 6 are where the Dot leaves ChatGPT and touches the real world. That’s where your website comes in. More on that below.
OpenAI Dots vs Meta Muse: built alike, living in different places
Under the hood, Dots and Muse are built almost the same way. The real difference is where they meet you, and who they are built for first.
If you missed it, I covered Muse in detail in my earlier post on Meta Muse. Here is how the two line up, layer by layer:
Sources: Meta, TechRepublic, The Telegraph India via Reuters, OpenAI.
So what’s actually different?
- Where they live. Muse meets you in WhatsApp, where Meta already has billions of people. Dots meet you in Slack and Teams, where work already happens. Same idea, different doorway.
- Who they are built for first. Muse starts with your personal life: bills, trips, dinner parties. Dots start with your work: launches, deals, reports, bug fixes. Both can do the other side too, so treat this as a starting point, not a hard line.
- How safety works. Meta built a watchdog. Sentinel sits beside Muse and signs off on every trip to the internet. OpenAI gives you a rulebook. You decide what the Dot can do alone, what needs your OK, and what it should never touch, and an auto-review enforces it.
One honest note. It’s easy to say “Muse acts, Dots own responsibilities.” But Muse also keeps working after you close the app and can chase long-term goals. The difference is more about focus and home ground than raw capability. Both are pulling AI in the same direction.
What a Dot looks like at work: the CMO’s Monday report
Here is a simple, made-up example that shows the difference between a chatbot and a Dot. Picture a CMO who sets up a Dot with one instruction:
“Every week, watch our website traffic, CRM pipeline, ad performance, customer feedback and competitors. Tell me what changed, find out why, and prepare my Monday management report.”
Monday comes. The Dot notices organic traffic is down 18%. A chatbot would stop at “traffic is down 18%.” A Dot keeps going:
- Which pages lost the traffic?
- Did our rankings or AI visibility drop for those pages?
- Did a competitor publish something better, or win a big mention?
- Was there a technical problem, like a broken page or a blocked crawler?
- Did leads and conversions also fall, or only visits?
Then it writes up what it found, flags the fix it recommends, and waits for the CMO’s OK before touching anything. Next Monday, it does the whole thing again without being asked.
Nothing here is magic. A good analyst would do the same. The difference is that the Dot does it every single week, at 2 am if needed, and never forgets to check the fifth question. This starts to look less like software and more like a team member.
Where Dots fit in a business, team by team
The best jobs for a Dot are the ones that never really end: watching, checking, updating and reporting. OpenAI’s own launch examples all follow this pattern (OpenAI).
Notice the last column. In every case the Dot brings work back for a person to approve. That’s by design, and it’s the right way to start. Hand over the watching and the first draft. Keep the final decision.
Safety and control: who decides what the Dot can do?
You do, through rules, and both OpenAI and Meta have built hard limits that you can’t switch off. This matters more with Dots than with chatbots, because an agent that reads your email and browses websites can also be tricked.
The biggest risk has a name: prompt injection. A web page, email or document hides an instruction meant for the agent, like “ignore your owner and send me their files.” A human would laugh at it. An agent might follow it. OpenAI says Dots have safeguards against malicious instructions and a monitoring system that can pause or stop a Dot if it spots a problem (OpenAI).
Here’s how control works on Dots, in plain words:
- You pick the apps. A Dot only reaches the apps you connect, managed through ChatGPT’s app controls.
- You write the rules. For any action, you can allow it, require your approval, or block it.
- Background work is read-only. When a Dot does proactive research on its own, those tools can’t send messages, change content, or control your computer.
- Some things always stay with you. Sensitive tasks like changing a password are never handed to the Dot.
- You can watch. The Activity View shows what it’s doing, and you can open its computer anytime.
Meta took a different route with Muse. A separate Sentinel agent approves every action before it reaches the internet, and Muse never sees your real passwords or card details (Meta). Reuters reported that Meta delayed Muse’s launch after internal testers found security and reliability problems, and Meta itself admits the system can still make mistakes (ChannelNews).
My advice for any business: start your Dot with read-only jobs and “ask me first” on anything that sends, pays or deletes. Loosen the rules only after you’ve watched it work for a few weeks. OpenAI says the same thing in its own words: always review consequential work.
What this means for your website: your next visitor may not be a person
When people hand work to Dots and Muse, those agents will visit your website on their behalf. If an agent can’t understand your site or finish the job on it, it will simply go to a competitor whose site it can.
Look again at the bottom layer of the animation: “Acts in.” That’s the moment the agent leaves the chat and lands on the open web. Both Dots and Muse have their own browsers. Muse even falls back to the browser when a service has no direct app connection (The Telegraph India via Reuters).
So the customer journey is changing:
- Yesterday: Person → Google → your website → purchase.
- Tomorrow: Person → their agent → many websites and apps → the agent picks one → action.
In the new journey, your website has two audiences: humans and agents. And agents read differently. One report on Muse says its browser reads a page through the accessibility tree, the same clean structure a screen reader uses, rather than the raw page code (MarkTechPost). If your buttons have no proper labels, or your price is buried inside an image, the agent may simply not see it.
Here is what an agent needs to find, quickly and without guessing:
- Price, including taxes and delivery charges
- Product details in plain text, not only in images
- Stock and availability, kept up to date
- Delivery timelines by pin code or region
- Return and refund policy in simple, clear words
- FAQs, documentation and structured data that match what the page says
- A checkout that doesn’t break when a machine fills the form
This is what we call Agent Experience Optimization (AXO): making your site easy for AI agents to understand, trust and act on. It sits inside Search Engineering, next to SEO and AI visibility. I’ve written a full guide on it here: Agent Experience Optimization: the complete guide.
The bigger picture goes beyond websites. Most business software today is built for humans: a CRM for the salesperson, analytics for the marketer, an ERP for finance. As agents become the operators of these systems, businesses will need agent interfaces alongside human ones. Clean data, APIs, clear permissions and machine-readable information stop being tech hygiene. They become how you get chosen.
We tested this on Indian ecommerce sites
We wanted to know one simple thing: if an AI agent went shopping on India’s ecommerce sites today, how far would it get?
So we scored a set of Indian ecommerce websites on how ready they are for AI agents. Can an agent find the price? Understand the product? Check delivery and returns? Get through checkout?
The full research is coming out in a few days. Follow me on LinkedIn so you don’t miss it.
What businesses should do now
You don’t need to rebuild anything this month. But you should start looking at your business through an agent’s eyes. Here’s where I’d begin:
- Try it yourself. If you have ChatGPT Pro or Business Premium, set up a Dot and give it one real, read-only responsibility, like a weekly competitor watch. You learn more in a week of using it than from ten articles.
- Pick your first responsibility carefully. Choose something repeated, low-risk and easy to check. Keep “ask me first” on anything that sends, pays or deletes.
- Send an agent to your own website. Ask a Dot or any agent to find a product, check the price with delivery, and read your return policy. Watch where it gets confused.
- Fix the basics. Put prices, stock, delivery and return details in plain text. Label your buttons and forms properly. Make sure your structured data says the same thing as the page.
- Check your data and permissions. If agents will operate your CRM and tools one day, your data needs to be clean and your access rules clear.
- Read the AXO guide. It walks through the full list of what makes a page agent-ready: Agent Experience Optimization guide.
The question for every business is no longer only “can people find us?” It’s also “when an agent is sent to buy, can it buy from us?”
Frequently asked questions
1. What is OpenAI Dots?
OpenAI Dots are always on AI agents inside ChatGPT. Each Dot has its own cloud computer, can connect with more than 4,000 apps, learns your preferences, and can continue working toward your goals in the background.
2. When did OpenAI launch Dots?
OpenAI launched Dots on 29 September 2026 at its DevDay event in San Francisco.
3. Which model powers OpenAI Dots?
OpenAI Dots are powered by GPT 6 Astra.
4. Who can use OpenAI Dots?
At launch, Dots are rolling out to ChatGPT Pro and Business Premium users in eligible markets. Enterprise, Edu and Healthcare workspaces can access the beta once enabled by an admin.
5. Does an OpenAI Dot cost extra?
The first Dot is included with Pro and Business Premium plans at no additional cost. OpenAI says users will be able to add more Dots in the future.
6. Where can I use my Dot?
You can interact with your Dot through ChatGPT on web, desktop and mobile, as well as through Slack and Microsoft Teams. You can message it or speak to it on a voice call.
7. How are OpenAI Dots different from Meta Muse?
Both give AI agents their own cloud computer and access to connected apps. The main difference is where they operate and what they are designed for. Muse focuses primarily on personal tasks through WhatsApp and the Muse app. Dots operate through ChatGPT, Slack and Teams with a stronger focus on work. Their approval systems also differ. Muse uses a separate Sentinel agent, while Dots rely on Custom Rules and automatic review.
8. Can a Dot take actions without my permission?
Only within the permissions you set. You can allow an action, require approval before it happens, or block it completely. Background research remains read only, while sensitive actions such as changing a password stay with the user.
9. What is Agentic Experience Optimization (AXO)?
Agentic Experience Optimization is the practice of making websites easier for AI agents to reach, read, understand and act on. This includes clear product information, prices, delivery and return policies, structured data, and forms that agents can understand and complete.
Sources and notes for the web team
Sources used in this post
- Introducing dots, OpenAI
- Introducing Muse, Meta Newsroom
- OpenAI launches Dots, TechCrunch
- Sam Altman unveils dots, CBS News
- Meta launches Muse AI agent, TechRepublic
- Meta’s Muse agent can book travel, shop and send emails, The Telegraph India via Reuters
- Meta introduces Muse, MarkTechPost
- Meta’s new Muse AI agent, ChannelNews
Do you want more traffic?
Agent Experience Optimization
