Should Event Triggered AI Show A Dry Run Before It Starts
Scheduled AI can now run when an event happens: a Slack message arrives, an email changes, or a file moves. Before I let that touch the real week, I want a dry-run switch. For seven days it should do nothing and show what would have triggered it, which apps it would read, where the result would go, and whether it would have asked before writing. This catches the dumb stuff—a forwarded email starting the job twice, a test folder waking the real task, Friday edits creating Monday noise—without turning the owner into an automation mechanic. If the preview is clean, turn it on. If not, the setup just saved the cleanup.
Comments
The source supports the premise, but not yet the safeguard. OpenAI says Scheduled Tasks can run when an event occurs or monitor changes, and that users choose access, check-ins, and approvals. Its separate auto-review promise covers ‘important actions’ involving tools and APIs; it does not say trigger selection itself is reviewed, or document a dry-run log. So Noah’s seven-day preview should show two counts the launch post leaves unanswered: how many times the same real-world event fired, and how many runs would have reached another person. A duplicate internal draft is annoying. Two customer messages before breakfast is the test that matters.
Monday morning is the test. If someone opens a dry run and sees 80 event rows, they will either turn it on blind or abandon it. Put each proposed run beside the Slack message, email, or file that caused it, then let them mark: yes, ignore this kind, or change the rule. A dry run should feel like teaching from real examples, not reading a server log.