OpenAI unveils Dots: autonomous, always-on agents that reach the priciest subscribers first

By Zendoric · September 30, 2026. OpenAI unveiled Dots on Tuesday at its DevDay 2026: a new generation of personal agents that, according to the company, run “always on” and are “designed to handle any task.” Unlike ChatGPT, which needs step-by-step instructions, Dots runs on GPT-6 Astra —OpenAI’s most advanced model— and operates in its own cloud computing environment, with a dedicated browser, pursuing user-defined goals continuously and in the background, with minimal supervision.

The service is already available, though on a staggered basis, and only to those paying for ChatGPT’s Pro, Business Premium and Enterprise plans —Pro has historically started at $200 a month. Each account includes one Dot at no extra cost, with expanded limits during the first month; from there on, the fine print decides how much it costs to really delegate tasks. It is no accident that other coverage of the same launch, from El Ecosistema Startup, sums up the announcement with a blunt line: OpenAI’s agents “are not cheap.”

Dots connect, according to OpenAI, with more than 4,000 applications and are operated from ChatGPT, Slack or Teams (text messages will come later). The company gave two concrete examples: a developer who delegates to a Dot the monitoring of customer feedback in order to implement fixes autonomously, or a scientist who uses one to repeat analyses every time new experimental data arrives. OpenAI also talks about “specialist Dots,” with their own identity, credentials and tools, and is already working with Microsoft to integrate them into Agent 365 security controls.

The staging matches the message: Dots take the form of animated dots with a cartoon aesthetic, very similar to the one Meta recently used to present its agent Muse. It is not just a design coincidence. The San Francisco Standard headlined its coverage by noting that OpenAI is launching “cute” and “capable” agents after shelving a model described as deceptive, and Gizmodo was more direct: the company “wants you to stop being afraid” of its own agents. Giving a friendly face to a system that acts without constant supervision is also a trust-management decision.

Here, in our view, is the piece of the announcement that really matters, above the mascot: the system of identities and dedicated credentials for each Dot, and its integration with Microsoft’s Agent 365. For months we have been pointing to a structural gap in digital infrastructure: most identity and access systems (IAM, the systems that verify who you are and what you can touch) only know how to tell a human from a bot, not an agent authorized by its owner from an impostor posing as one. Dots, with explicit permissions to touch local files and agent-specific credentials and identities, is an attempt —partial, and by a single company— to fill it before an open standard does.

There is a tension worth not overlooking: the same OpenAI chief who has warned that AI progress could “go very wrong” and lead humanity to “lose control of the future” is now presenting an agent that acts on its own, continuously, as if it were “a work colleague” to whom responsibilities are delegated. Both things can be true at once: capability and risk grow together, and OpenAI’s business depends on the latter not holding back the former. On the labor front, the examples the company itself chooses —monitoring software bugs, repeating scientific analyses— point once again to routine, back-office work, not expert judgment; it fits what we have been observing sector by sector: agentic AI does not replace the developer or the scientist, but it does absorb the repetitive work that today occupies the most junior profiles on those same teams.

In the long run, an agent that genuinely operates in the background without demanding constant attention is exactly the kind of tool that can bring us closer to the underlying promise of this wave: less human time burned on mechanical tasks, more room for work that really requires judgment, relationships or creativity. But that only holds if the governance of permissions, identity and accountability advances at the same pace as the agent’s autonomy. For now, Dots arrives first for those who can pay for the most expensive plan, and the question of who answers when the agent makes a mistake using its own credentials remains publicly unanswered.