← Back to articlesHarnesses and AI coding

How do you stop AI bug fixes from causing more bugs?

Reproduce, diagnose, make a minimal change and retest (checking again using the same steps).

The short answer

Preserve exact errors and triggers. Avoid repeated reinstallations or rewrites without new evidence.

Try it in three steps

  • Record page, action, expected result and actual result.
  • Ask for the leading hypothesis and a test.
  • Retest the original steps and neighboring behavior.

A practice prompt

Before you start

This is a bug-report exercise, not a confirmed diagnosis.

Practice material (fictional)

Symptom: clicking the menu immediately after loading fails, but clicking after 2 seconds works; desktop is unaffected.

Replace placeholders in the prompt with this material, or append the material if there are no placeholders. Reading time excludes doing the exercise.

Reproduce the first-click failure and distinguish missing handlers, obstruction and readiness. Investigate before changing three suspected causes.

What to look for (not a recorded test result)

Initialization or obstruction are hypotheses requiring observation. Retest the immediate click rather than waiting two seconds and declaring success.

If it goes wrong

If a different path works after the fix, still test the original; do not change the criterion.

Check your result

The original error is resolved, adjacent behavior works and scope is clear.

References and scope

The links below support the concepts or product navigation. The steps and exercises are our teaching suggestions, not official certification or measured customer outcomes. Product interfaces and availability may change. References checked on 2026-10-17.

Start with one task

Describe the current process and its slowest step. The initial conversation is free.

Let’s talk ↗
Need help? See small-scale implementation services ↗

Keep reading