Lesson 5 of 5, 11 min with the exercise

Keep it running after launch

Stop your automation breaking quietly.

The usual failure is silence: a login expires or an API changes, and nobody notices for two weeks. Plan for this before you launch.

• Name one owner.
• Send an alert when a run fails, and when no runs happen at a time you'd expect some.
• Write down how to do the task by hand if it stops.
• Check quality on a schedule, and track the monthly cost.

Model updates can change behaviour too. Keep your test cases, and rerun them when you change the model or the prompt.

See it in practice

Runbook example: "Owner: Dina. If there's no report by 9am Monday, check the run log, then the API login. If it's not fixed by 11am, use the manual steps in this doc and tell the team." It's short, but someone covering for Dina can follow it.

Common mistake: Treating launch as the finish line. Without an owner and alerts, the first you'll hear of a failure may come from a customer.