ChatGPT Work has browser capability that can read, click, and type on webpages, including authenticated sites. ChatGPT also has scheduled tasks that can run later. What is missing is reliable integration between those capabilities.
A customer should be able to authorize an autonomous task to perform a browser action when it becomes available: open the authenticated page at the scheduled time, detect the enabled control, click the specified button, and verify whether the action succeeded.
This is needed for any limited-window registration, reservation, appointment, application, submission, or inventory action. A reminder telling the customer to press the button is not autonomous execution.
The workflow should:
- Confirm before scheduling that browser access and authentication will be available during the autonomous run.
- Open the authorized page at the trigger time.
- Detect when the requested action becomes available.
- Click the authorized button.
- Verify and preserve the result or confirmation.
- Immediately report the exact blocker if execution fails.
If the scheduled execution environment cannot perform the requested browser action, ChatGPT must say that before accepting the task. It should not create an apparently actionable autonomous task that will only produce information, a reminder, or no action at all.
This is both a missing capability and a capability-disclosure problem. Time-sensitive execution is one of the clearest reasons to use an autonomous agent.
OpenAI’s Browser documentation describes ChatGPT Work as capable of clicking and typing on websites, using authenticated sites, and completing appointment-booking workflows: