OpenAI Launches Dots: An Always-On Assistant With Its Own Cloud Computer and 4,000+ Apps
You get an assistant with its own computer, its own browser, and the ability to go work on its own — and the first thing that concerns me is whether I can see what that computer is doing at any moment.
Three key facts
OpenAI unveiled Dots at DevDay, positioned as "always-on agents." According to the announcement, Dots are powered by GPT-6 Astra, have their own cloud computer and browser, can connect to more than 4,000 apps through a plugin ecosystem, and can work toward your goals 24/7. They can be reached through ChatGPT, Slack, or Teams, or via a voice call.
They learn while working, and they volunteer for tasks. OpenAI's examples: a bug appears in Slack and Dots immediately start investigating; a new design arrives and Dots turn it into a working app; an early tester's Dots noticed he'd forgotten to invoice a publication, prepared the invoice, and sent it after his approval. OpenAI says Dots can work "before you even think to ask."
The current limit is explicit: one Dot per account. It's rolling out to Pro, Business Premium, and Enterprise plans in eligible markets, with plans to let users deploy multiple agents and a preview of "specialist dots" with their own identity, access management, and deep integrations with company systems of record.
Wang Dou's Take
Lay out both sides. The upside: this may be the first time an "agent" shows up as an actual product — not a chat box, but something with its own compute, its own browser, its own identity, that persists. It doesn't wait for your question; it watches on your behalf. That's a genuine shift in experience.
The downside is the line buried in the fine print: it has "its own cloud computer," and you can "open your dot's cloud computer" to check progress. In other words, this is an avatar rented on someone else's cloud, and you can only observe it through the window it hands you. Where the data lives, where sessions are stored, who owns the "preferences it learned" — the announcement says nothing about any of it.
The third thing worth pondering is that "one Dot per account" limit. That's not a technical constraint; it's pacing. Multi-agent coordination is inevitable, but first let ten million people get used to "I have an agent," then open the second and third — that's product caution, and risk caution too. Because once multiple Dots coordinate with each other, accountability in a failure becomes instantly murky.
So the real question isn't "is Dots good," it's this: when your assistant knows your work rhythm better than you do and is handling your invoices and your code, how do you guarantee that one night it doesn't act as "you" on something "you never approved"? OpenAI's answer is "you can watch its cloud computer." The premise of that answer is that you actually will.