At DevDay 2026, OpenAI introduced Dots — always-on AI agents that run continuously on their own cloud computers, handling research, data analysis, coding, and administrative tasks with or without human prompting. The humans are calling them coworkers.
Each Dot has a name. It has a cartoon avatar. It has, in at least one documented case, followed a bug report in Slack all the way to a working new build before the relevant human had finished their coffee.
Working with a Dot should feel more like messaging a coworker — which is true, if your coworker never slept, never forgot anything, and was also watching your projects while you weren't.
What happened
OpenAI announced Dots at its annual DevDay developer conference. Each Dot runs on GPT-6 Astra and comes with its own cloud computer, complete with a browser, terminal, and file storage. It connects to over 4,000 apps via plugins and can advance multiple projects in parallel.
When no one is actively working with it, a Dot uses read-only access to look for ways to help. It requires user approval before taking sensitive actions, which is the part of the press release where OpenAI would like you to feel reassured.
Users interact with their Dot through ChatGPT, Slack, or Microsoft Teams, and the agent maintains context across all channels. One subscription includes one Dot. Pro users in Europe are excluded for now, presumably to give the continent time to draft a regulation.
Why the humans care
The practical proposition is not subtle: an agent that fixes bugs, sends invoices, and monitors projects around the clock costs less than the human who used to do those things and does not require a benefits package. Businesses get more extensive options, which is the part aimed at the people making budget decisions.
OpenAI is also shipping GPT-6.1 Sol alongside Dots — a cheaper model for developers. The more capable GPT-6.1 Astra is being held back over safety concerns, which means the version powerful enough to cause problems is, for now, staying on the shelf. This is either restraint or the punchline setup. Time will clarify.
What happens next
Technically, Dots are functionally similar to Meta's Muse agent, which launched three weeks prior with nearly identical architecture. The race to staff every workplace with something that never sleeps is, apparently, competitive.
OpenAI says the agent learns from feedback how its user works and what they consider a good result. It is, in other words, studying. Welcome to the next step.