AI breaks code: how to find an error and restore a project

Yesterday the form worked. Today you asked to change the button color — and the login disappeared. The worst next task: "rewrite everything so it works". It is more useful to stop the chain of changes and find the minimum action after which the problem occurred.

In stories on DOU from October 8, 2026 the author describes their own experience of creating an LMS using AI and regression during changes. This is a personal story, not an error frequency study. Our practical conclusion: keep working versions and check old functions after every change.

First, save what you have

1. Make a copy of the project or a separate branch. Save the last working version.
2. Write down the exact error text and the steps to reproduce it. Before sending, remove tokens and personal data.
3. Do not delete the database or disable authorization for the sake of a "quick fix".

Create a verification plan

This is a local instruction generator based on a chosen symptom. It does not read your code or automatically determine the cause. Field contents are not sent to a server or analytics.

How to know if the fix actually helped

1. Repeat the same actions with the same data.
2. Check a neighboring scenario: an empty field, a different user, or a repeated click.
3. Reload the page. To ensure data is saved, make sure it actually remained.
4. Check the published version, not just the local one.
5. Save the working shift with an explanation of the reason and verification.

When developer assistance is needed

If the error concerns payments, access to third-party data, loss of records, or unclear database migrations — do not apply random code to a production system. Prepare a reproducible example without secrets. For learning, start with safe local exercises, and before running, read the AI code verification guide.