# Debugging with an Assistant — AI Coding Assistants: Use Them Well and Review Them Hard

Source: https://www.geekswithgeeks.com/en/ai-coding-assistants/flow-debug

> Combine stack traces, hypotheses and experiments to find and fix bugs faster.

## Hypothesis, experiment, result

Give the assistant the **stack trace, the code around it and what you expected versus what happened**. Ask for several possible causes ranked by likelihood, then test them one by one with a print, a log or a failing test. Feed each result back. Avoid accepting a "fix" you do not understand: a patch that makes the symptom vanish may hide the real bug.

## A bug an assistant often misses

A mutable default argument is shared between calls. I ran this: the buggy version returns the same list twice, the fixed version gives independent lists.

```python
def add_item(x, bucket=[]):          # buggy: one shared list
    bucket.append(x); return bucket
print(add_item(1), add_item(2))

def add_item_ok(x, bucket=None):
    bucket = [] if bucket is None else bucket
    bucket.append(x); return bucket
print(add_item_ok(1), add_item_ok(2))
```

Output:

```
[1, 2] [1, 2]
[1] [2]
```

## Write a failing test first

Reproduce the bug as a test before fixing it. The test proves the fix worked and stops the bug coming back, and an agent can use it to confirm its own patch.

**Quiz:** Why avoid accepting a fix you do not understand?

- [ ] It makes the editor slower
- [ ] Fixes are always wrong
- [ ] Git forbids it
- [x] It may hide the real bug while removing the symptom

*Answer:* It may hide the real bug while removing the symptom. Understanding the cause is how you know the problem is truly solved.