Dots, OpenAI's agent that watches your email and calendar unprompted, launches proactive AI at $100 a month

By Fast Company · October 9, 2026.

OpenAI unveiled Dots on September 29 at the keynote of its DevDay conference in San Francisco. It is an agent that is given access to the user's calendar and email and that, without anyone asking, detects problems and raises the alarm. Columnist Harry McCracken, of Fast Company's Plugged In newsletter, tells the story with a case of his own: Dots pointed out that the date of a talk he had in his calendar did not match the one in a confirmation email. The organizers had moved the event to another day and had forgotten to tell him; without the agent's alert, he would have shown up to an empty room.

For now Dots is only available to those paying for the most expensive tiers of ChatGPT, starting at $100 a month, according to the article. The company plans to extend it to more users once it has refined it and resolved its cost model. The service runs on GPT-6 Astra, the model OpenAI had announced earlier that same September, and assigns each user a Linux computer in the cloud for the agent to use as a working environment.

The piece is part of a newsletter and the available material is brief —a personal anecdote plus the product announcement— but the technical move it describes is significant: it is the difference between an assistant that answers and an agent that watches. Until now, the narrative around "agentic" AI was largely aspirational: flashy demos, use cases confined to coding or web browsing. A system that cross-references two sources of personal data —email and calendar— and detects an inconsistency without being asked is a demonstration of everyday usefulness, not a showcase.

It also exposes OpenAI's usual deployment pattern: the most advanced capability reaches whoever pays the most first. A threshold of $100 a month is not a symbolic barrier, it is a line separating those who can delegate their administrative management to an AI from those who still do it by hand. It is exactly the kind of task —reconciling dates, cross-checking confirmations, catching scheduling errors— that in our sector analysis of business administration we flagged as the most exposed to automation: the job does not disappear overnight, but the value of someone who does nothing beyond that routine checking work does.

There is also an underlying question that the article does not raise but that should not be glossed over: an agent with permanent read access to email and calendar, operating from a cloud computer assigned to each user, is also a new attack surface. If that access is compromised —through a failure of the system itself or through manipulation of the agent via malicious instructions inserted into an email— the same mechanism that warns of a date change could be used to extract sensitive information. The promise of proactivity and the risk of autonomy are two sides of the same coin, and it deserves the same attention we pay to the risk of agentic fraud in other contexts.

Our reading is that this kind of product marks better than any benchmark where AI's real frontier lies: it is not just generating text or code, it is managing a person's administrative life with judgment of its own. In the short term, that will remain expensive and exclusive, and will reinforce the gap between those who can afford an artificial "chief of staff" and those who cannot. But the price trajectory of generative AI in recent years —from the exclusivity of the first paid models to the almost free availability of equivalent capabilities a short time later— suggests that this function will get cheaper and become widespread. The underlying horizon remains the same one we defend at Zendoric: the more routine work these agents absorb, the more time is freed up for people to devote to what really matters to them, provided the transition does not leave behind those who cannot pay the entry fee.