OpenAI Just Launched Always-On AI Agents (Dots Explained)

Vídeo em Inglês, publicado pelo canal WesGPT.

Video transcript

Transcript from available captions; it may contain errors.

At OpenAI's Dev Day 2026, they introduced dots. And in this video, I'm going to explain exactly what this new feature is and how we can use it to improve our workflows. All right, so the big catchphrase for OpenAI's dots is that it works while you sleep. So it's ChatGPT's always-on agents compared to the normal chatbots where you have to be online at your computer typing to it in order for it to work. So when you log off your computer, your dot doesn't. It's always working throughout the day while you sleep. It's constantly checking your apps and your integrations. It's researching. It's writing. It's doing your pre-approved tasks. Now, for dots, OpenAI is making them these little characters. They're remarkably capable, always-on agents built to handle everything. Their brain is GPT-6 Astra, and they're saying it learns your goals and standards from feedback over time. So that's another thing. These dots are constantly changing, and they're adapting to you specifically. So it's its own computer. Each dot has a cloud computer and a browser of its own. You can open it at any time and watch it work. So picture the computer that you're on right now. Each of the dots you create is going to have one computer for itself. And this computer is in the cloud. It has tools. It has browsers. And that's what allows it to be on 24/7. So right out of the gate, it can connect to 4,000-plus apps, and it connects through plugins on OpenAI. How do dots work? So step one, they learn. Your goals, your standards, how you like things done. It picks this up from every correction that you give it. So the idea is that the more you use it, the better it performs. Step two is it's work on its own computer with your connected apps. It will work when you're away, and it will also do proactive [snorts] research based on predicting tasks that you will have in the future. So, it's going to stop making assumptions and just working. It's going to proactively research and then come to you for approval. And then the third step, review, it brings you the results, you approve, correct, or redirect it, and then the loop starts again. So, you can find dots in multiple different places. You can find it at ChatGPT, it's desktop, web, and mobile. It connects through Slack. I've seen a lot of AIs lately connecting through Slack. So, Slack is taking big leaps there in the conversational AI agents. You can connect with Microsoft Teams. You can make voice calls to it when you need to talk things out. That's pretty cool. I saw in the Dev Day video, the girl actually picks up her phone and calls her dot, and the screen on her phone is like she's calling a real person. And then they're going to have texting, which is coming soon. So, this is a visual of it working. It looks like inside Codex, but the point is it carries context across every channel that you connect it to, and it can message you with progress, questions, or decisions that need you. So, wherever you live, if it's either on the web in ChatGPT, in Slack, or in Microsoft Teams, you can connect to it. So, this is an example of a night with your dot. So, I'm going to click play here, and this is an example of after you've gone to sleep. So, you go to sleep, your work is finished for the day, and at night, it's proactively researching. So, it read three new emails from Hollow Pine Press. Then it noticed that invoice number 231 for your August article was never sent. So at 12:31 a.m. it drafted the invoice from your template. That's $1,200 net 30 days. At 12:33 it checked the custom rules and one of the custom rules that you set and we'll talk about this security and privacy later, that in order to send invoices it requires your approval. So at 6:58 a.m. when you wake up, your dot sends you an approval notice and you can either approve or edit what it did. So background work is read only. It can look, draft, and plan, but it can't send or do anything. Your rules decide what waits for your approval. And then this is how it is set up your dot. So at the start you name it, for example, Juniper. Then you connect what apps you want your dot to work in, for example, Gmail, Google Calendar Slack GitHub Notion Google Drive, Stripe, Linear. Remember there are 4,000 apps you can connect to. Then you say hello and let it introduce itself. And the important part is correcting it when it does things wrong. And then you set all of the rules. So for example, you're going to allow it to read the calendar and inbox, require approval before sending emails and invoices, block it completely from making purchases. You could also allow that, which would be interesting. And change my passwords. You want to say only you can do that. So the rollout of dots is today and it's sending out to pro and business premium accounts in the eligible markets. I'm in Ontario, Canada and I have not gotten the dots feature yet. So let me know in the comments below if you have. And then for the cost, dots are included in the plans. Chatting with it doesn't count towards your usage, but as soon as you send it to work, that usage does count. So that is how they're doing the pricing. There are three controls to know in dots. So, we have custom rules. This is what we talked about earlier. So, allow an action, block it outright. These dots start with built-in rules for when to act and when to ask you things. There's auto review. So, it checks any action that could touch your accounts or share information against your rules. I think they pre-built some security checks like changing a password that always stays with you. Like, can you imagine if your dot changed your password overnight and you had no way to access any of your accounts anymore? That'd be crazy. And then activity view. And this is where you can follow everything it does including its background work and then you can redirect it whenever you want. So, while you're away, your dot is read-only. And this is a cool little security feature. So, all of its background work and all of your connected apps, they're restricted to read-only. It can't send messages, it can't change content in your apps, and it can't control your browser or your computer. Your dot's computer and your computer, they stay separate unless you choose to connect them. And a fair warning that OpenAI's monitoring system, they can stop your dot if it detects a safety concern. So, I think they're trying to get you always kind of on your phone and always looking at your dot's messages. Like, if a call comes through or a message comes through in Slack that needs your approval with your dot, you're going to have to approve, deny, and redirect it. So, clearly this isn't fully automatic. There's still a huge human-in-the-loop component. But, the point is you're taking AI everywhere with you on the go. In the presentation, they showed five jobs that could use dots. First one is a developer turning customer feedback into scoped, tested fixes, and brings you PRs. so those are pull requests with videos of the changes. This example is a launch lead. When the scope changes, it rewrites the story and revises the launch materials for your review. Another example is a scientist. It reruns analysis as new data arrives. It updates the figures for you and flags what needs your eyes only. This example is a sales lead. Your dot checks requirements against product docs and then builds a proof of concept and updates the proposal. And how about a creator? So, your dot gets a new transcript in, it finds clip moments, write show notes, and drafts social media posts for your approval. This is a visual of what that could look like. So, the developer example, Iggy scopes a sidebar fix, then hands over the iOS beta checklist for review, and then the creator, Jojo, drafts the post. Now, those names, Iggy and Jojo, those are your dots. It shows the notes and the clip ideas, then applies one edit across all of them. So, think of it as a personal assistant that works in the cloud. Now, the caveats to this new feature, obviously, it can still make mistakes. OpenAI tells you to always review consequential work. The approval step is the point of your dot learning. It's not just a formality. They're trying to bake that into the process. It's not available for everyone yet. Like I said, I don't have it yet and I'm on a pro account. And then that texting feature is coming soon. Having more than one dot and their pricing aren't announced yet. And that first month of introducing dots, you're going to get extended limits and usage, but that's going to shrink over time, just like they do with every good model. All right, that is a full review of the new dots feature released at DevDay 2026 by OpenAI. I really want to hear your thoughts on this. Do you think this can be useful? How do you think this can be improved? Do you think this is similar to anything else we already have? To me, it seems a little similar to Claude Codes Cloud Sessions, which is actually something I haven't been using as much as I should, but I definitely want to hear your thoughts. So, please drop a comment on this video. And then one more quick thing before we go, I'm dropping a lot of free tools on my personal website Spreng.com. All of these are completely free. I got a really cool one coming soon called Audio Slides. But right now we have an AI video editor. This video that you're watching right now was created fully with this tool. So, it definitely works and it creates awesome-looking videos. We also have a low-fi video maker and Say Screen, which turns a website into a narrated demo video using AI. So, come check us out. And if you like this video, I put two more on the screen right now. Both have been catered to your personal YouTube watch history. Give one of them a click for me, give it a watch, and I'll see you in there. Peace.

Assistir ao vídeo original no YouTube

Assuntos: Agentes de IA, Dots. Publicado no YouTube em 2026-09-29.

Seleção automática por assunto e idioma. O vídeo pertence ao canal de origem.

About specialist dots and dots data use

Alongside the primary dot each user gets, OpenAI previewed specialist dots on September 29, 2026, the day dots were introduced. While a primary dot works on its user's behalf, a specialist dot takes on dedicated responsibilities inside an organization: the company sets it up with its own identity, credentials and access to the systems it needs, along with IT-provisioned hardware and deep integrations with the company's systems of record. OpenAI says it is building on lessons from early internal testing in procurement, invoice processing, email marketing, customer support and commercial contracting.

OpenAI is starting with focused enterprise pilots, in which its engineering teams work directly with organizations to define each dot's responsibilities, the tools it can use and how people review and approve its work. It is also working with Microsoft to integrate specialist dots with the governance and security controls of Microsoft Agent 365. Enterprise users, including Edu and Healthcare workspaces, can try the dots beta when their workspace admin enables it. Content from ChatGPT Business, Enterprise and Edu workspaces is not used to improve OpenAI's models by default, and OpenAI says it does not train directly on proactive research or a dot's notes to itself.

Fact sheet

• Product: Specialist dots (OpenAI)
• Previewed: September 29, 2026
• Setup: Own identity, credentials, system access and IT-provisioned hardware
• Early internal testing: Procurement, invoice processing, email marketing, customer support and commercial contracting
• Availability: Focused enterprise pilots
• Microsoft integration: Agent 365 governance and security controls
• Enterprise beta: Includes Edu and Healthcare, enabled by the workspace admin
• Workspace data: Not used to improve models by default

Read also

• 🚨 OpenAI Just Launched “Dots” — AI Agents That ACT Before You Ask
• OpenAI Dots: AI That Keeps Working While You're Away
• OPENAI JUST CHANGED AI AGENTS — What Are Dots?

---

Sources and references: OpenAI – Introducing dots · video by the WesGPT channel on YouTube.