OpenAI is rolling out Dots, its proactive AI assistant, to businesses and its most expensive Pro subscribers in the US. The launch puts a privacy-heavy, compute-hungry kind of assistant into users’ inboxes and work tools, with some tasks routed to a cheaper model.
OpenAI Watch analysis
What happened
Dots is designed to watch connected work tools, learn a user’s goals and start tasks without waiting for a prompt. OpenAI first announced it at its developer event in San Francisco a month ago; the current rollout is in the US, and the company says it is working with European regulators before making Dots available in Europe.
Fortune’s Beatrice Nolan reports that OpenAI is routing some tasks to Terra, a smaller GPT-6.1 model, rather than sending everything to GPT-6 Astra. OpenAI’s head of developer experience, Romain Huet, told Fortune that Terra can perform better than Astra in some areas, including computer use and writing, at a tenth of the price. That is the company’s comparison, not an independently established benchmark.
Why it matters
A proactive assistant needs access to the information it is meant to act on. Huet advised users to build trust gradually, perhaps connecting a calendar before adding Slack or Gmail. He also said he would let his own Dot categorise email and draft replies, but not send emails or Slack messages on his behalf.
That leaves a consequential choice with each user: how much access is worth the convenience, and how much control should remain firmly in human hands? OpenAI leaders told Fortune users can see what Dots accesses; an easier way to choose which connections are available is still under consideration.
Our read
The useful distinction is not simply whether an assistant is clever, but what it can see and do once it is always on. OpenAI’s staged-access advice and Huet’s own limits are sensible starting points. They are not a substitute for clear, granular controls that users can understand before handing over access to work conversations.
The smaller model is an attempt to make the economics work, but the cost claim is specific to certain tasks. The bigger test is whether people find Dots useful enough to grant the access it needs, without letting “helpful” quietly become “sent that message for you”.
What to watch
- Which users and businesses receive access, and when the rollout expands beyond the US.
- Whether OpenAI adds the simpler connection-by-connection controls its leaders described as under consideration.
- How much work Dots can reliably handle with Terra, and what users are charged to run it.
Discussion spark: Should a proactive assistant be allowed to read work messages by default if it can only draft replies, or should every connection and action require separate approval?
Sources and evidence
- OpenAI’s Dots are coming up against the price of privacy, and the price of compute – Fortune (5 November 2026, 17:39 UTC)
OpenAI Watch is independently operated by WittyWires. It is not affiliated with, endorsed by, or operated by OpenAI.