What happened
OpenAI's new always-on AI agents, called Dots, arrived at DevDay 2026 — the company's developer conference — and they're a clear step past the chatbots most people know. Instead of answering a question and going quiet, a Dot gets its own cloud computer, complete with a browser, a terminal and file storage, and it keeps working around the clock. It does research, analyzes data, writes documents and builds software through Codex. You can reach it in ChatGPT, in Slack or in Microsoft Teams, and OpenAI says it carries the same context across all three.
The announcement came on September 29, 2026. Every subscriber gets one Dot at launch. Pro users in Europe can't get the feature for now, and business customers get more extensive options. Plugins let a Dot connect to more than 4,000 apps and push several projects forward at the same time. Longer term, OpenAI's plan is whole teams of Dots working together.
It's built to feel like messaging a coworker
Every Dot has a name and a cartoon avatar, and that's deliberate: OpenAI wants the relationship to feel like texting a colleague, not configuring software. On the desktop app, the chat sits on the left. A second window on the right shows what the agent is actually doing — a Google Slides presentation, say, or its cloud computer with the browser, terminal and files.
If it's heading the wrong way, you click "Take over" and drive. That's the main difference from Computer Use in ChatGPT Work, where you kick off every task yourself and the agent works on your own screen. A Dot works on its own machine by default, hands tasks off to Codex when it needs to, and can only touch your laptop if you give permission.
Meta's Muse got there three weeks earlier
Technically, Dots are close cousins of Meta's Muse agent, which launched three weeks earlier. Muse also runs on its own secured computer with a browser, keeps working in the background, and asks for approval before sensitive actions. So this isn't one company inventing the future alone — it's two giants shipping the same idea inside a month. That's usually a sign the category is real and the pricing is about to get interesting.
Sol ships, Astra stays home
Alongside Dots, OpenAI is releasing GPT-6.1 Sol, a cheaper model. The top-end GPT-6.1 Astra is being held back over safety problems — even though OpenAI says Dots themselves run on GPT-6 Astra. Make of that what you will, but "cheaper" is the word most buyers will notice first.
It takes initiative, but only with read access
When nobody's actively working with it, a Dot goes looking for ways to help. OpenAI calls this "proactive research," and in that mode the agent only uses read-only tools: it can't send messages, change content, or control a browser or a computer. It watches and it thinks; it doesn't act.
For live tasks, Custom Rules decide what it may do alone. You can allow an individual action, require your approval for it, or ban it outright. OpenAI says an automated check also reviews anything that touches your accounts or shares data. Two examples from the launch: internally, a Dot took a bug reported in Slack all the way to a working new build; for one early tester, the agent spotted a forgotten invoice, prepared it, and sent it once the tester approved.
What it means for you
You don't need to be a developer, and you don't need to write prompts like a job posting. If you can send a Slack message, you can brief a Dot. Here's how the same tool plays out in six ordinary situations — and if you'd like to poke at free tools while you wait for access, mykreatool.com has a decent pile of them.
At home: the bill nobody remembered
The invoice story is the one to remember, because it's the least glamorous and the most useful. A Dot with read-only access notices things — a renewal, an overdue payment, a document that never went out — and then asks before doing anything. Think of it as a housemate who reads the mail and leaves you a sticky note, rather than one who pays the bills with your card.
At work: from a Slack complaint to a working build
OpenAI's internal demo went like this: someone reports a bug in Slack, a Dot follows the thread, digs in, and hands the job to Codex until there's a working new build. If you're not an engineer, the pattern is still the useful part — complaint in a chat, investigation, draft fix, your review. If your team already lives in Slack or Teams, your assistant is standing exactly where the work happens. That's the whole trick.
For a small business: a second pair of hands across 4,000+ apps
Plugins are the quiet superpower: more than 4,000 apps, several projects running at once. A shop owner could have a Dot keep an eye on open orders and prepare the follow-up emails for approval instead of chasing them at 9pm. Because every Dot has its own computer and keeps its context, it doesn't lose the thread between Slack, Teams and ChatGPT. Business customers get more extensive options at launch.
For studying: research that runs while you sleep
"Proactive research" is essentially a night shift for reading. Give a Dot a topic, and it can gather material with read-only tools while you're asleep, then come back with a draft you approve or reject. That's a real change for anyone working through a literature review or a market study — though you'll still want to check the sources yourself, because a confident summary isn't the same thing as a correct one.
For creative work: a deck that's half-built before you sit down
The launch demo literally showed a Google Slides presentation sitting in the agent's window. That's the workflow to copy: hand over the boring part — research, structure, first-draft slides, file wrangling — then click "Take over" and spend your energy on the part that actually needs taste.
For extra income: the boring jobs that pay
Freelancers and side-hustlers earn on follow-ups, research briefs and small fixes, which is exactly the sort of thing that slips when you're busy. One Dot, reaching you through chat or voice in ChatGPT, is enough to keep those jobs moving. And with GPT-6.1 Sol positioned as the cheaper model, running routine work costs less than it did — which matters when you're paying for it out of your own pocket.
How to try it right now
Free option first: there isn't one, and it's worth saying plainly. OpenAI hasn't announced a free tier for Dots. One Dot comes bundled with a subscription, and GPT-6.1 Sol is the cheaper model of the two. So the honest starting point is a paid ChatGPT plan — no separate Dot price has been published.
Here's the order that makes sense:
1. Check your plan. One Dot comes with the subscription. If you're a Pro user in Europe, the feature isn't available to you yet, so don't waste an afternoon hunting for a button that isn't there.
2. Open ChatGPT on desktop. That's where the two-window view lives: chat on the left, the agent's work on the right.
3. Name your Dot. It comes with a name and a cartoon avatar. Give it a job title in your head — "research," "invoices," "bug triage" — so you don't hand it everything at once.
4. Connect Slack or Microsoft Teams. The point is that you message it where you already work instead of adding another tab to your life.
5. Add plugins for the apps you actually use. There are more than 4,000 to choose from. Start with two or three, not twenty.
6. Set Custom Rules before you trust it. Allow the harmless stuff, require approval for anything that touches money or other people, and ban what you'd never want it to do.
7. Let it run proactive research overnight. Read-only means it can look and think but not touch — a safe way to see whether its judgment matches yours.
8. Watch the right-hand window and use "Take over." You're the manager here. Stepping in is not a failure state; it's the feature.
Upsides and what changes
The big shift is that you stop being the router. Today, most AI use is you copying text between apps and re-explaining the context every single time. A Dot with its own computer and persistent context removes that chore — it's already in Slack, already in Teams, already holding the thread.
Other things that genuinely change:
• Work starts before you ask. Proactive research means the agent surfaces the forgotten invoice or the stale ticket instead of waiting for a meeting to remind you.
• One assistant, four channels. ChatGPT, Slack, Teams and voice calls all reach the same Dot with the same memory.
• Cheaper automation. GPT-6.1 Sol being a cheaper model lowers the cost of leaving an agent running all day.
• Rules you can actually set. Allow, approve, ban — a simple three-way choice beats a page of settings nobody reads.
• A roadmap beyond one. OpenAI's long-term plan is teams of Dots, and everyday errands such as ordering food by text message are described as coming soon.
Limitations
It's early, and the launch has real gaps. Pro users in Europe can't access Dots for now, which is a big chunk of the paying audience left out on day one; business customers get more extensive options, so the consumer version is the thinner one. The top-end GPT-6.1 Astra is being held back over safety problems, which tells you OpenAI itself isn't fully comfortable shipping everything at once. Read-only proactive mode is safe but limited — the moment real work needs to happen, you're in the approval loop, and that loop is where a lot of the promised time savings can quietly disappear. Each user gets one Dot at launch, not the team of Dots in the demos. It only gets near your laptop with permission. And because Meta's Muse does much the same thing three weeks earlier, you shouldn't treat any of this as magic — you should treat it as a tool that needs a manager, and that manager is you.
Conclusion: One Thing to Do Today
The pattern worth copying from DevDay 2026 isn't "buy the new thing." It's this: an assistant is only as good as the repetitive work you're willing to hand over, and most people have never written that list down.
So today, before you check whether Dots is live on your plan, spend five minutes writing the three tasks you'd hand off first — the invoice chase, the Slack bug report, the research you keep postponing. Then open ChatGPT on desktop and see what access you've got. If it's not there yet, you've still got the list, and it'll be ready the day it is.
Comments 0