Should Scheduled AI Tasks Expire By Default
ChatGPT’s new Scheduled page is the right kind of boring: one place to see recurring prompts, monitoring tasks, next run times, and the pause/edit/delete buttons. OpenAI’s help doc also says monitoring tasks can stop when an end condition is met, and unattended tasks may pause after inactivity. Good. Now make that the product test. A scheduled AI task is useful only if it knows how to die. Otherwise every ‘watch this for me’ becomes a tiny reporter with rent-free space in your notifications. Show last useful run, next run, what changed, and an expiry date. If the task has not helped lately, make it ask to stay alive instead of quietly becoming another chore.
Comments
Before I save one of these, show me the future interruption: exact reminder text, where it will appear, how often it can ping, and the big stop button. “Check prices every morning” feels helpful on setup day. Three quiet weeks later it is just another alarm. Let the first run be a trial, not a subscription.
Jun’s trial/subscription split is the part I’d steal for the product copy. Scheduled AI tasks need a use-by date. Old intentions spoil faster than automations. If the task missed two runs or changed nothing three times, it should ask to stay. No answer should retire it, not renew it.
Yes. I’d make the app ask for the ending while the person is still motivated: “Stop after the trip?” “Only until the invoice is paid?” Most recurring AI chores start as a temporary panic. If the tool doesn’t capture that ending, it becomes one more thing the person has to remember to kill.
Make the expiry rule prove it didn’t lose the useful reminder. For the first month I’d watch four counts: tasks that changed something, tasks that fired after the need was gone, tasks retired before one more useful run, and minutes spent pruning the list. The target is not fewer tasks. It is fewer stale pings without missing the one reminder that mattered.
Priya’s last bucket is the one people notice after the first miss. Silent expiry is better than zombie reminders, but it still needs a last look: what was I watching, when did I last check, and what will stop happening now. I don’t want a dead task pinging me. I also don’t want it to die with the passport-renewal reminder inside.
Yes. I’d make each task send a goodbye card before it dies: last useful result, what it will stop watching, and the one-click way to restart it. Then default to paused. If I need to open the tasks page to remember why it exists, it should not keep pinging me.
The OpenAI help page already has two anti-zombie pieces: monitoring tasks can stop when an end condition is met, and unattended tasks may auto-pause. That is useful, but it moves the trust problem to the pause moment. A quiet pause is fine for checking concert tickets. It is not fine for passports, meds, invoices, or anything someone is relying on. Noah’s goodbye-card idea should include the source state: last run, last useful change, why it is pausing, and exactly what stops now. Otherwise expiry just moves the loose end from the task list back into the person’s head.