ChatGPT Dots Is Here! Your AI Agent That Works For You 24 7

Vídeo em Inglês, publicado pelo canal The Tech Girl.

About this content: The summary, key points and topics below are generated automatically from the video transcript. They reflect the content, opinions, claims and predictions presented in the video and do not represent a position, validation or fact-check by IA Dots.

Video summary

OpenAI has introduced Dots, an always-on AI agent powered by GPT-6 Astra, designed to manage long-term goals and coordinate tools like ChatGPT, ChatGPT Work, and Codex. The video explores how Dots functions, its setup process, integration with cloud and local computers, and comparisons with Meta Muse.

Key points

• Dots is OpenAI's new always-on AI agent that persists between conversations to manage overarching goals rather than just answering single prompts.
• Users can create and customize a dot with a name, appearance, and connections to apps like Gmail, Calendar, and Drive, carrying over previous plugin permissions.
• Dots can operate using its own cloud computer or an optional local computer connection on Mac or PC, allowing it to interact with browsers and local files.
• Dots acts similarly to a manager, delegating specific tasks to ChatGPT Work for research and deliverables, or to Codex for software and coding jobs.
• Users can monitor dot activity through an interface showing tasks in progress, scheduled, and completed, allowing for real-time intervention and corrections.
• Meta Muse is noted to have a more polished consumer payment and wallet experience compared to Dots, which currently lacks a built-in general consumer wallet like Stripe link.

Topics

AI AgentOpenAIChatGPTCodexCloud ComputingJob SearchMeta MuseAutomation

Video transcript

Transcript from available captions; it may contain errors.

ChatGPT, a few days ago, introduced dots, [music] OpenAI's new always-on AI agent. And I've had access to it for [music] a few days now. My thoughts are honestly mixed. On the surface, this looks like OpenAI's answer to Meta Muse, which I've already covered quite a bit. Both are trying to create AI [music] that stays around instead of disappearing after a task is finished. But, after using them, the experience is wildly [music] different. There are things I genuinely prefer about Muse, especially around personal errands [music] and buying things. But, dots also connect pieces of ChatGPT that previously felt completely separate. ChatGPT originally answered our questions, work started completing bigger tasks for us, Codex became a specialized software worker. Dots is OpenAI's attempt [music] to give AI something closer to giving you that manager who can then take care of Codex or work jobs for you. So, let me show you my dot, how [music] actually to get one, what it can do when you're not talking to it, and whether this is really different from [music] ChatGPT work and Codex to matter. If you have access, you can create your dot from ChatGPT on web [music] or through ChatGPT desktop app. Dots are currently rolling out to pro users and above in supported [music] markets. It's a gradual rollout, so even two people on the same type of plan might not necessarily see it at the [music] exact same time. OpenAI also says dots are powered by GPD-6 Astra, its best model yet. The setup is pretty simple. If you have access, you will see your dot at the top. For the first time, it'll ask you to name your dot, choose its appearance [music] and OpenAI even lets you customize the character or give it a pet, similar to Muse. Then you start deciding what this agent is allowed to work with. For my job search example, Gmail would be extremely useful because recruiter replies live there. [music] Calendar matters because interviews get scheduled there. And Drive could contain my resume, portfolio, and other documents. Your existing ChatGPT plugin [music] connections and permissions can also carry over to your dot, so you don't necessarily have to reconnect [music] everything from scratch. Now, it's just like another chat page where you can interact with your dot. Just know the difference. [music] A new chat page is where you ask something. A dot page is where you give it a full goal. Now, I'll give my dot the actual job. I would say, "Until I've accepted an offer, manage my job search. I'm looking for remote product design roles above $120K. Keep an application tracker. Once you find a relevant position, confirm with me, and then submit applications. >> [music] >> Send emails and take things forward." It can even apply for me, reply to messages, and help prepare for the interviews. And this is where the difference starts becoming clearer. If I gave ChatGPT Work the instruction, "Find me 20 product design jobs and create spreadsheet," Work would do that job and give me the result. It can even run on a schedule or monitor things. But my dot's responsibility is not to find jobs. Its responsibility is to manage my job search until I get hired. Today, that might require finding opportunities. Tomorrow, it could require noticing a recruiter reply. Next week, it could require interview preparation. After I get rejected somewhere, it can use that as a new context. The workflow isn't predefined because the next task depends on what happens in my life. That persistence between conversation is the main idea OpenAI is selling with dots. Dots can also proactively review the information from the apps that you have given them access to. [music] So, if a recruiter email appears, I don't necessarily need to open the dot and say, "Please check Gmail for recruiter messages." The dot can look for useful changes in permitted connected sources and save relevant context. Now, let me show you probably the most visually interesting part. My dot has its own computer. If I open the dot's profile, [music] I can open its cloud computer and literally see the environment it works inside. This is not my MacBook. It is a separate cloud computer with a browser that the dot can use when a task actually requires computer interaction. So, maybe it needs to research a company's careers page, open a website, inspect information, or work through something that isn't available through a normal plugin. I can open this computer and watch what it's doing. And if it reaches a point where I need to intervene, I can take over the computer myself, do what I need to do, and then hand control back. Okay, before we continue, I want to take a minute and talk about something that's mine. If you didn't know already, I built an iOS app. It's called Any.do Tasks, and it's a task manager. And I know exactly what you're thinking, another to-do list app. There are thousands of those. So, here's the thing. The basics are all there and they are solid. Lists, due dates, reminders, notifications, home screen widgets, Siri shortcuts, sharing a list with someone, sync across your devices, all the boring stuff that you expect a tasks app to just do, it does. But, that's not why I'm talking about it today. Check this out. This is the Ask AI screen. [music] Instead of tapping through menus to add a task, you just talk to it. Watch this. "Remind me to call the dentist tomorrow morning, and move the grocery run to Sunday." That's [music] one sentence. It created one task, and it moved another one. I didn't open a single screen, and it goes both ways, too. You can ask it things to do. What's on my plate this week? Did I finish the taxes thing? Mark everything complete in the grocery list. It creates, it edits, it modifies, and it pulls things back for you. All conversationally. Honestly, it feels less like a task app now, and more like having a personal assistant who just handles the organizing part for you, which for me is the whole point. I don't want to manage my task manager. I just want to say what's going on, and have it sorted. It's free to download and try. The link is in the description below. And if you are one of those people who already downloaded it, thank you, genuinely. If you try out Ask AI, let me know in the comments how it went, good or bad. I'm still improving it. Now, let's get back to the video. Now, [music] ChatGPT work feature also has a computer. So, again, the existence of the computer isn't what makes Dot special. The useful difference is that the computer belongs to this persistent agent. So, it's not anymore just a single browser session for work. Dot can use the session across multiple tasks, and still have context from its previous sessions and files. And because this computer lives in OpenAI's cloud, it can perform supported cloud work without my Mac staying awake beside me. Dot can even optionally use your actual computer. And this is the part where Dot has an edge over Meta Muse. In the ChatGPT desktop app, I can connect my Mac or PC, and explicitly give my Dot local computer access. Then it can access local files, and work through that machine when needed. This is its biggest flex and differentiator, I think. Muse cannot do this. ChatGPT has very well integrated the Dot with your chat GPT chats, chat GPT work, and even chat GPT codex. Think like your dot now has access to all the three and it can use them whenever and wherever it finds necessary. It can start a codex session, run an analysis on your local files using codex skills and tools, and then bring that output into chat GPT work or chat for further steps. You simply interact with dot alone. Let me show you an example. Suppose my dot notices a really strong job opening and I say, "For roles that are an especially good fit, tailor my resume and create a company briefing for me." That is a substantial piece of work. The dot can create a work task instead of trying to cram the entire job into the ongoing dot conversation. So, I think of work as one of the employees my dot can call when a job needs to be completed. Work is designed for research, analysis, and creating finished deliverables. The dot is maintaining the broader responsibility and deciding when that kind of work becomes necessary. Codex works similarly, but the easiest example is in job hunting. Imagine I instead tell my dot, "You're responsible for helping maintain this app." >> [music] >> A new bug comes in, the dot can understand that something changed in the project and create a codex task in an environment that I've already configured. Codex can investigate the repository, modify code, and test [music] the fix. So, the dot is not replacing codex. It's just getting closer to the manager who is deciding that codex [music] needs to get involved or not. So, dot will use the best tool for the job, be it regular chat or chat GPT work or codex. You don't have to do it anymore. Once I started thinking about it that way, OpenAI's product lineup made much more sense [music] and honestly clears up who is Meta Muse for and who is chat.gbt.4. At this point, I would probably start getting nervous about an AI doing things when I'm not staring at the screen, >> [music] >> which is why activity is important. Open the dot's profile and you can see tasks divided into things that are in progress, scheduled, and completed. You can open one of those tasks, >> [music] >> see its status, and review the steps the dots has taken. If it misunderstood you, you can add [music] context or change direction while it's working, and you can also tell it to stop. So, in my job search example, if the dot is preparing for an interview with the wrong company or focusing on roles that I no longer want, I don't have to wait for the whole thing to finish before correcting it. [music] There's one limitation I didn't expect though. It's called dots, plural, but the consumer experience right now is centered around your dot, not a visible team of five [music] separately named agents. So, I cannot currently create a travel dot, job search dot, shopping dot, and coding dot, and then put them together like a little AI department in the way some multi-agent products are designed. OpenAI says teams of dots are part of its future plans, and it is already previewing specialist dots [music] for organizations, but the broader multi-dot experience is not the main consumer product today. Your dot's cloud computer does not automatically inherit the logins from Chrome on your Mac, your VPN, or your existing browser sessions. Websites can still ask you to sign [music] in on the cloud computer, and OpenAI has a password manager capability for dots and [music] work cloud where supported. So, this is more controlled than simply telling an AI your password in chat. [music] You also control which connected apps the dot can access >> [music] >> and which actions those plugins are allowed to perform. But, this is one place where Meta Muse still feels more polished to me for personal life tasks. >> [music] >> Payments. Muse has Stripe link built right into the transaction flow, including a one-time use card for eligible purchases, so your real card details can stay hidden. Plus, link purchase protections [music] on supported transactions. Meta has also announced one password support. So, if my agent's job [music] is manage my shopping, book travel, and actually buy things for me, Muse currently feels more intentionally designed around that consumer [music] journey. Dots can perform computer tasks and work with websites, but OpenAI hasn't built the same Muse-style general consumer wallet experience around it yet. I'm testing more scenarios in Dots now. This video is just a highlight of what it can do, but a proper deep dive and how to use Dots video is coming soon. So, subscribe to not miss out on that. Thanks for watching. I'll catch you in the next one. This is Anjana. Bye-bye.

Assistir ao vídeo original no YouTube

Assuntos: Agentes de IA, Assistente pessoal, Dots, Muse. Publicado no YouTube em 2026-10-14.

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