Reading AI Output and Debugging
Explore how to evaluate AI-generated code effectively by reading error messages, spotting silent failures, and pinpointing which layer—frontend, backend, or database—is causing issues. Learn practical debugging methods to test user flows thoroughly, report precise problems to AI tools for fixes, and decide when to rollback changes to maintain stable app development.
We'll cover the following...
- What should you check after every change?
- What do error messages actually tell you?
- Why does the AI sometimes make things up?
- What about bugs that produce no error at all?
- How do you figure out which layer is broken?
- How do you use the AI to fix its own mistakes?
- When should you roll back instead of fixing forward?
- What comes next?
You asked the AI to add a booking confirmation email to PawPals. It said “Done! I’ve added email confirmation functionality to the booking flow.” You clicked through the app. The booking form still works. The confirmation page still appears. Everything looks fine.
Two days later, a friend tests the app and tells you they never received a confirmation email. You go back and check. There is no email service connected. There is no API key. The AI wrote code that formats and “sends” an email, but the sending part goes nowhere. The code runs without errors. The app does not crash. It just quietly does nothing.
The AI was not lying. It built what it described. But “done” meant “I wrote the code,” not “I verified it works end to end.” That distinction is yours to catch. This lesson covers what to check after every round, how to read error messages, why the AI sometimes confidently produces things that do not exist, and what to do when something breaks with no error at all.
What should you check after every change?
The preview looks right, the page loads, and the button you asked for is there. That is not enough.
After every round of changes, run through four checks. They take less than a minute and catch problems that would otherwise compound for three more rounds before you notice them.
Walk the full user flow. Not just the part you changed. Start from the homepage, click through to the booking form, fill it in, confirm, check the confirmation page. Changes in one area can quietly break something in another, and the AI will not warn you.
Test whether data survives a refresh. Enter something into a form, submit it, then reload the page. If the data is gone, either the app is not saving to a database (which is expected if you have not set one up yet) or the save is broken.
Check mobile. Resize your browser window or open the app on your phone. Layout that looks fine at full width can collapse on smaller screens.
Look for regressions. A regression is when something that was working before stops working after a change you did not intend to affect it. The navigation bar vanishes. A button that used to open a modal now does nothing. You find them by ...