OpenAI's always-on Dots agents can read while you're away, but not write
On 29th Sep'26, OpenAI launched Dots at DevDay: persistent agents that run on their own cloud computers, each with its own browser, connect to more than 4,000 apps and work across ChatGPT, Slack and Teams. They roll out to ChatGPT Pro and Business Premium users in eligible markets, with one dot included at no extra cost, while Enterprise, Edu and Healthcare workspaces get a beta only once an administrator enables it. When the user is away, a dot can do proactive research through connected apps in read-only mode, and any action that writes needs approval under rules the user sets. Password changes and other account-security tasks always stay with the user.
What it actually means
The headline is the always-on agent, but the useful part is where OpenAI drew the line: reading is allowed unattended and writing is not. That is the permission model most teams building agents have been improvising, and a lab with this much usage has now picked a default, so expect customers to ask why your agent does not behave the same way.
If you ship an agent that touches customer systems, split its tools into read and write before anything else, and make the write side wait for a person unless a rule says otherwise. It costs you some of the magic of an agent that just does it, but it is the difference between an agent that flags a problem overnight and one that emails a client about it.
Read is safe to automate; write needs a person.
4,000+apps a single dot can connect to
Source: The Next Web · Thursday 1 October 2026