i broke down OpenAI’s new “Dots”

OpenAI just dropped Dots, and honestly, I think the interesting part isn't any one feature.

It's the way they stitched a bunch of things we've already seen in AI agents into one persistent system.

I went through the official Dots announcement and the launch video, and tried to break down what actually sits underneath it.

Here's how I see it.

1. the main thing: it's always there

A normal AI is basically:

you ask → AI responds → done

Dots is trying to change that into:

you give it a goal → it works → keeps working → comes back when there's something useful

OpenAI describes Dots as always-on agents that can work toward your goals 24/7 and handle multiple projects at the same time.

That's a pretty big shift.

You're not just chatting with an AI anymore.

You're basically giving it a job.

---

2. it gets its own computer

This is probably one of the coolest parts.

Every Dot gets its own cloud computer and browser.

So instead of the model just telling you what to do, the agent actually has an environment where it can do the work.

It can use its browser, connected apps, files, etc.

And you can apparently open up its computer and actually see what it's doing.

So the architecture starts looking more like:

AI model + computer + tools

instead of just:

AI model + chat box

---

3. it remembers how you work

This is another huge part.

Dots aren't supposed to forget everything after every task.

OpenAI says they learn your goals, preferences, standards and what "good" looks like to you through your interactions and feedback.

So over time, ideally:

you give it feedback

↓

it learns your preferences

↓

next task needs less explanation

↓

eventually it starts doing things more like you

That's where this starts becoming genuinely interesting for things like coding, research, writing, content creation, etc.

---

4. you can literally talk to it

Dots aren't limited to a chat window.

You can interact with them through ChatGPT, Slack and Teams, and OpenAI also lets you voice call your Dot.

So voice becomes another interface into the same agent.

Something like:

voice → STT → agent → tools → work → response → TTS

Which makes the whole thing feel much more like having an assistant you can just talk to.

---

5. then you give it access to your apps

This is where the agent becomes actually useful.

OpenAI says Dots can connect to 4,000+ apps through its plugin ecosystem.

So instead of the model existing in isolation, it gets access to the software you already use.

Think:

reasoning → tools → apps → actions

That's basically the difference between an AI that can talk about doing something and an AI that can potentially actually do it.

---

6. the browser is basically another tool

Because the Dot has its own cloud computer and browser, browser automation becomes part of the system too.

And this is one of the things I find really interesting.

Imagine telling your agent:

"go through these websites, collect the information, compare everything, and prepare the result."

Instead of giving you instructions, the goal is for the agent to actually go and do it.

That's the direction a lot of agent systems are heading toward.

---

7. and it doesn't really care where you talk to it

This part is easy to miss.

You can start something in ChatGPT, continue through Slack, talk to it through voice, etc.

The Dot is supposed to carry the context across those interfaces.

So the Dot is the actual persistent entity.

ChatGPT / Slack / Teams / voice are basically interfaces into it.

And that's a pretty interesting mental model.

---

so what is a Dot actually made of?

My simplified breakdown would be:

persistent agent

• memory
• cloud computer
• browser
• connected apps
• voice
• background execution
• permissions / approvals

Put all of those together and you get something that's much closer to a persistent digital worker than a chatbot.

And OpenAI is already talking about going further with specialist Dots for organizations, where different agents can have their own identity, credentials, tools and responsibilities.

So the progression could basically be:

one AI assistant

→ one persistent agent

→ multiple specialized agents

→ a whole team of agents

That's the part of Dots I find the most interesting.

Not because any individual component is completely new, but because the pieces are finally being packaged into one persistent system.

---

architecture

Architecture above is my simplified breakdown based on OpenAI's public Dots announcement, not an official OpenAI architecture diagram.

Source: OpenAI — Introducing dots