Design and test an automation that handles a stable task without allowing AI output to create an unreviewed external result.
Select an internal task with a stable trigger, reversible action and named reviewer.
A first automation should create an internal record or draft, not send a customer message or make a financial decision. Keep the trigger and finish condition observable.
A fictional support form with consent set to yes creates one internal case. AI may choose an allowed label but cannot answer the customer.
Name the trigger, approved fields, allowed AI output, reviewer, external actions prohibited in the first test and stop rule.
Turn this task into an automation boundary: [task]. Return trigger approved fields AI output allowed labels reviewer prohibited actions and stop rule. Mark missing policy or permission as NEEDS REVIEW.
No. It teaches a tested specification and first build method. You choose a permitted platform and connection.
No. The course starts with a fixed automation and one bounded AI step.
Only when the chosen tool, account and policy allow it. The first exercises use synthetic data.
