Tepesoft Systems

What should you test before releasing an AI-built app?

Start with the paths that can expose customer data, charge money, or lose work. Use this checklist to turn an AI-built prototype into a release you can review and recover.

Write down the user journeys that must work

List the three to five things a customer must be able to finish on launch day. Give each journey a starting state, a visible result, and a failure case. For a booking app, test a successful booking, two attempts to book the same slot, and a retry after the confirmation request times out.

Check each journey with a fresh account and with realistic existing data. Include the narrowest screen or slowest connection you support and verify the result from the user's point of view.

Verify permission and data boundaries on the server

Use staging accounts with different roles and confirm each account sees only the records and actions it is entitled to use. Exercise authorization through the server, since hiding a button only changes the screen. Review the configuration for committed secrets and verify that production credentials and test credentials are separate.

If the app calls an AI service, document the data the request sends and what happens when the service times out or returns unusable output. Require a meaningful review step before automation can send messages, publish content, or make changes a user cannot easily undo.

Leave enough evidence to make a release decision

Record the build or commit tested and the result of each critical check. Give unresolved issues an owner and document which ones block launch. Keep a repeatable check for every defect that previously stopped a critical journey.

Before launch, rehearse the recovery steps in staging. Identify the last known good version, who can activate it, and how a rollback would affect any data migration. Decide how the team will detect a failed rollout, such as a jump in rejected sign-ins or failed bookings.

References