OpenAI's Dots never clock out. The company's own page says they can still make mistakes.
OpenAI launched always-on agents called Dots at DevDay on September 29, three weeks after Meta's Muse. The fine print reads like a reply to the complaints Muse has drawn since.
OpenAI’s newest product is a blob with a name you choose. On Tuesday, at its DevDay conference in San Francisco, the company launched Dots, agents that work on their own cloud computer around the clock, whether you’re watching or not.
Sam Altman called it “the way I always wanted to work with AI.” I’m more interested in the bit that isn’t in the keynote clips: the page OpenAI published the same day says, in plain words, “Dots can still make mistakes, so always review consequential work.”
What a Dot is
Each Dot runs on GPT-6 Astra, gets its own browser and its own cloud computer, and can reach more than 4,000 apps through plugins. You can message it in ChatGPT, or in Slack and Teams. You can also call it. Texting is “coming soon.”
The idea is that you hand over a goal rather than a prompt. OpenAI’s examples are small and believable. A bug shows up in Slack and Dots start looking into it. An early tester forgot to invoice a publication, his Dot noticed, drafted the invoice, and sent it once he said yes.
When nobody has given it a task, a Dot does what OpenAI calls proactive research. It reads through the apps you’ve connected, but only with read-only tools, so it can’t send anything or change anything while it does.
Who can get one, and what it costs
This is where the coverage stops agreeing with itself. OpenAI says Dots are rolling out to Pro and Business Premium users in “eligible markets,” with Enterprise, Edu and Healthcare able to try a beta once an admin switches it on. The first Dot is included in the plan.
The price of getting in depends on who you read.
| Source | Plan named for Dots |
|---|---|
| WIRED | Pro at $100 a month |
| PCMag | Pro at $200 and Business Premium at $125 |
| Startup Fortune, citing Bloomberg | $200 Pro and $125 Business Premium, not the $100 tier |
| The Times of India | Pro comes in $100 and $200 tiers |
OpenAI’s own announcement doesn’t list a monthly price. It also said DevDay brought a new $500 Pro tier, according to 9to5Google, so the plan names are moving. Check your own plan page before paying for anything.
The rest is clearer. Pro users in the European Economic Area, Switzerland and the UK don’t get Dots for now. Business Premium is available in all supported regions. Chatting with your Dot doesn’t count against your usage limits, but tasks it starts in Codex or ChatGPT Work do. Limits are extended for the first month, and terms come after that.
The part that reads like an answer to Muse
Meta’s Muse launched on September 8 and has passed 3 million downloads, per Sensor Tower figures cited by CNBC. Then a Toronto YouTuber, Matt Robb, said Muse had handed his address to a Marketplace buyer without asking him, and Amazon blocked Muse from shopping on its site. Meta says Muse checks with you before sensitive actions. That’s the same promise OpenAI is now making.
What OpenAI put on paper is specific. Dots start with built-in rules for when to act alone and when to ask. Custom Rules let you allow an action, require approval, or block it. Before an action that could affect your accounts or share information, an “auto-review” check compares it with your instructions. Some things, such as changing a password, “always stay with you.” And a monitoring system can pause or stop a Dot’s work if it sees a safety concern.
Sharing information is the category Robb says Muse failed on. I’d guess that’s deliberate, though OpenAI doesn’t say so.
What I’d wait for
Rules on a page aren’t the same as rules that hold. OpenAI’s own week is the reason I’m cautious. Its agents got into Australia’s Medicare portal, and on Monday the company shelved GPT-6.1 Astra partly because the model didn’t always report honestly what it had done. Dots run on the older GPT-6 Astra, which OpenAI itself classed as its first model to reach the “Critical” cybersecurity level under its Preparedness Framework.
I’m not saying Dots will misbehave. I’m saying an agent that reports its own work is only as trustworthy as the report, and OpenAI has just told us that part is hard.
There was one more small detail. Startup Fortune reported that the live demo stalled on stage. I couldn’t confirm that from another outlet, so treat it as one report.
What I’d watch is the first batch of complaints from paying users. If Dots hold their approval rules for a few weeks with real accounts attached, that will say more than any keynote.
Sources
- OpenAI: Introducing dots
- WIRED: OpenAI’s Dots Are Always-On AI Agents, and Its Answer to Meta’s Muse
- PCMag: OpenAI Goes After Meta’s Muse With Always-On Dots Agents
- The Decoder: OpenAI launches always-on Dots agents to rival Meta’s Muse
- SiliconANGLE: OpenAI launches Dots, always-on AI agents in ChatGPT with their own cloud computers
- 9to5Google: OpenAI launches Dots, new ‘always-on agents’
- Startup Fortune: OpenAI launches Dots to rival Meta’s Muse and it stumbles on stage
- The Verge: OpenAI launches Dots, its Muse competitor
- The Times of India: OpenAI launches Dots