OpenAI’s announcement post on September 29, 2026 describes Dots as “remarkably capable, always-on agents built to handle everything” and calls them “a whole new way to work with AI—one that gets to know what matters to you.” That’s a big claim wrapped in friendly language, and my first reaction was less about the capability and more about the phrasing. Something that gets to know what matters to you, and never turns off. That’s the part worth slowing down on.
Dots landed at DevDay 2026, OpenAI’s developer conference, held Tuesday in San Francisco. They’re depicted as cute blobs. You can personalize them, hand them multi-step projects, and they go work on those projects on your behalf. Mashable’s coverage from the event featured a yellow blob mascot named Alfred up on a screen. Behind the cartoon exterior sits GPT-6 Astra, and according to SiliconANGLE’s reporting, each one gets its own cloud computer.
What “always-on” actually changes
If you’ve used ChatGPT, you know the rhythm: you ask, it answers, the conversation ends. You’re the one who starts things. Nothing happens when you close the tab.
Dots break that rhythm. They’re built to constantly crawl the web to complete tasks for you. Reuters described them as agents that chase user goals across apps on their own. So the mental model shifts from tool to something closer to a standing arrangement. You describe an outcome, and work continues while you’re asleep, in a meeting, or thinking about something else entirely.
For non-technical folks, here’s the simplest way I can frame the difference:
- A chatbot waits for your question and gives you an answer.
- An agent takes a goal and works through the steps to reach it.
- An always-on agent keeps doing that without being re-prompted each time.
That third row is new territory for most people. Not because the technology is hard to understand, but because we don’t have habits for it yet. We know how to use a search box. We don’t really know how to manage a coworker who never sleeps and never asks for a break.
The cute blob is doing work
I want to give OpenAI credit for the design choice, and also be honest about what it accomplishes. Making agents look like personalizable blobs solves a real problem. Abstract software that acts on your behalf is hard to picture. A blob with a name is easy to picture. You can point at it, talk about it, assign it something.
The flip side is that friendly visuals make it easy to forget what’s underneath. A blob named Alfred crawling the web on your behalf is still software making decisions about what to read, what to trust, and what to do next. The cartoon doesn’t reduce that. It just makes it feel smaller than it is. That’s not a criticism of the design so much as a note to keep in your pocket.
The enterprise angle tells you who this is for
The detail that tells me the most is the Microsoft integration. Dots work with Microsoft’s enterprise governance and security controls in Agent 365, which means businesses can manage them through tools they already have.
That’s not a consumer feature. That’s an answer to the question every IT department asks before anything new gets approved: who can see what, who approves what, and how do we shut it off. Building that in at launch signals OpenAI expects companies, not just individuals, to be running these. Reuters framed the launch as deepening OpenAI’s push into the enterprise market, and both Reuters and WIRED positioned Dots against Meta’s Muse. So this is a competitive move in a market where the customers are companies with procurement processes.
Who can try them
Dots are rolling out in ChatGPT to Pro and Business Premium subscribers in eligible markets, with a beta for enterprise users. If you’re on a free plan, this one isn’t for you yet. That’s typical for a launch like this, and honestly it’s reasonable. Agents that run continuously and operate their own cloud computers cost real money to keep running.
What I’d watch for
The interesting question isn’t whether Dots are capable. It’s whether people build good habits around them. Always-on means you need a sense of what you’ve delegated, what’s currently in motion, and when to check the work. That’s a new skill, and most of us don’t have it.
My suggestion, if you get access: start with one small project you’d genuinely be fine having done imperfectly. Watch how it goes. Learn the failure modes on something low-stakes before you hand over anything that matters. The blobs are friendly, but the work they’re doing is real.
đź•’ Published: