The Debugger Who Forgot How to Debug
You used to read a stack trace and know where to look. Now you paste it into AI before you have finished reading it.
A senior engineer told us recently:
"I used to be able to read a stack trace and know approximately where the problem was within 30 seconds. Now I paste it into Claude before I have finished reading the first line. And here is the thing: I know I am doing it. I watch myself do it. I cannot stop."
This is what we are calling debugger drift — the slow, invisible migration of your debugging instincts to AI.
What debugging actually requires
Good debugging is not about finding the error. The error message tells you where the error is. Good debugging is about understanding why the error happened — what sequence of decisions led to this moment, what assumption was wrong, what interaction between systems created this particular failure.
That understanding requires exactly the kind of deep model-building that AI assistance bypasses. When you debug with AI, you get the answer. When you debug without AI, you get the model.
The 20-minute rule
Here is the practice that recovers debugging instinct:
The 20-Minute Debug Rule
Before you paste any error into AI: spend 20 minutes with it alone. Read it. Think about it. Write down where you think the problem is and why. Then — and only then — use AI.
The 20 minutes is not about whether you solve it yourself. It is about building the habit of engaging deeply with the problem before accepting a shortcut.
What you write down before using AI is the real data: what did you notice? What did you try? Where were you uncertain? That record — not whether you solved it — is the measurement of whether your debugging instincts are still active.
Why this is different from other skill erosion
Debugging instinct is particularly fragile because it compounds in the background. Unlike writing a function from scratch — where the absence is obvious — debugger drift happens gradually. You stop noticing you are using AI for errors you used to solve instantly.
Until one day you get an error in an environment without AI access, and you realize how much you have lost.
What helps
- The 20-minute rule above
- No-AI debugging sessions: one problem per week, no AI, no Stack Overflow — just you and the error
- Write down your debugging process: when you solve something, note what clue led you to the root cause
Debugging instinct is rebuildable. But like all recovery, it requires choosing to feel the discomfort of not knowing — before the shortcut.