How we handle user-reported bugs, UI changes, and other requests

1. Understand the request
• Read the user’s message carefully.
• Clarify what they expected vs. what actually happened.
• Collect reproduction steps, affected pages/URLs, screenshots, and device/browser details if needed.

2. Validate the issue
• Check the user’s website to confirm the problem is real.
• Do a quick site impact check to understand how widespread the issue is.

3. If the issue is valid: fix it safely
• Locate the cause in the codebase:
• Application code
• CSS and layout
• User-injected custom code
• Implement the smallest safe change that solves the problem.
• Test on a demo/staging environment first when possible.
• Verify on the live site after release.
• Ensure the fix does not break existing pages, components, or layouts.

4. If the issue is not valid
• Respond clearly with what was verified.
• Explain why it is not a bug (or why it is expected behavior).
• Provide guidance for the user to adjust settings/content to achieve their goal.

When the problem is an error in the codebase

1. Capture the error
• Identify the error message and where it appears (console, server logs, build output).

2. Analyze and fix
• Use AI to help locate where the error originates and propose a fix.
• Apply the fix in code.

3. Verify
• Manually test the site end-to-end.
• Add a regression check so the same error does not happen again.

How we define and build a new feature with AI support

1. Define the spec
• Write clear requirements and acceptance criteria.
• Discuss with product owners and get approval.

2. Research the best approach
• Review existing code patterns.
• Read relevant docs and references.
• Decide on the simplest approach that fits the spec.

3. Build with human-in-the-loop oversight
• Use AI for planning, architecture ideas, and code drafts.
• Engineers review every change before merging.
• If something is wrong, correct it and iterate.

4. Test and iterate until stable
• Test against the spec.
• Fix issues found during testing.
• Repeat until the feature works without regressions.