dots remembers instructions on behavior and checks before acting
OpenAI on Nov. 12 (Nov. 13 Japan time) announced an update to its always-on AI agent "dots" through ChatGPT's official account on X. The company said it made replies faster, cut down on jargon and improved the agent's ability to follow through on what it is asked to do. dots also now follows "custom instructions" set by the user. The post was published at 5:04:24 a.m. Nov. 13 Japan time.
Your dot now has… ⚡️ Faster replies 💬 Less jargon 🫡 Better follow-through And, they can now follow your custom instructions. Tell your dot in the thread things like how to behave or which actions need your go-ahead. “always talk in lowercase.” “Show me a draft before sending any emails.” You can also set custom instructions and rules in Customize > Behavior on mobile, or through your dot’s menu on web and desktop.
According to the post, users can tell dots in a thread how it should behave and which actions need their go-ahead before it carries them out. As examples, it gave an instruction on wording ("always talk in lowercase.") and an instruction to check before acting ("Show me a draft before sending any emails."). It said the settings are in "Customize > Behavior" on mobile and in the dot's menu on web and desktop.
dots is an agent that handles tasks on the user's behalf. Since the start of November, OpenAI has added replies from iPhone notifications and connections to multiple local computers in quick succession. This time, in addition to adding a feature, the company named the speed and clarity of its exchanges as areas of improvement.
Instructions are kept separate from ChatGPT's own settings
OpenAI's update log lists "Set custom instructions for your dot" under the week of Nov. 9–13. It says users can tell their dot how they would like it to respond and work with them, and can either ask it in conversation to update the instructions or edit them in "Customize → Behavior."
According to OpenAI's help documentation, custom instructions are set for each dot and are kept separate from ChatGPT's own custom instructions. Users can set preferences such as writing style, how often it checks in with progress, and when it should ask before taking an action. If users tell dots in conversation which instructions they want it to keep following, dots remembers them.
In the mobile app, where the "Behavior" option is available, users can review and edit the instructions under "Customize > Behavior" in the sidebar. If "Behavior" does not appear, the help documentation advises updating the app or using the dot's chat on desktop or web.
"Custom rules" decide whether actions may proceed
The help documentation also describes "custom rules," which set how specific actions are handled. For each action, users choose one of four options: "Take action without asking," "Take action when you say so," "Ask before taking action" and "Hand off to you." As an example, it suggested asking before sending messages to customers and handing off the deletion of shared project files to the user.
The help documentation separates the roles: preferences such as writing style go in custom instructions, while custom rules decide when the dot may take an action. It says rules are instructions that dots tries to follow and that dots can make mistakes. Rules cannot grant access to an app or computer, and they cannot remove built-in safety requirements or required confirmations such as approval before using a saved login.
No figures on the speed gains or the plans covered
The post gave no figures showing how much faster replies became or how much jargon was reduced. It also did not explain how the improvement in follow-through was measured. The post includes a video, but ORIENS did not use its content as a basis for this article.
Neither the post nor the update log states which plans or regions this update covers, or when it will be available. According to OpenAI's help documentation, dots itself is rolling out gradually to eligible accounts on the Pro plan and on Business Premium, a plan for businesses. On Pro plans, it is available to users aged 18 and older outside the European Economic Area, the United Kingdom and Switzerland. On Enterprise, Edu and Healthcare plans, it is available in eligible workspaces.
