I cannot debug without pasting the error into AI
Keep pasting, and add one step in front. Read the error, find the line it points to, and write one sentence: "I think this is wrong because..." Give yourself five minutes to test that guess. Then paste the error with your guess, and ask what you missed.
What is going on
Red text appears, and your hands copy it before you have read it. The fix comes back, it works, and you move on. It is quick, and for a typo or a missing import it is perfectly sensible.
Debugging is also where you learn how a system really behaves, and it seems to be the skill that fades first. In a trial with 52 mostly junior developers (Shen and Tamkin, Anthropic, 2026, a preprint that is not peer reviewed), the group that used an AI assistant scored about 50% on a later skills quiz against 67% for the group without, and debugging questions showed the largest gap. The sample is small and the quiz came straight after the task, so it says nothing about the long term.
What to do this week
Read the error out loud
The type, the message and the first line of your own code in the trace. The quick fix is often already in that text.
Write your guess
One sentence in a comment or a notes file: what I think is wrong. A wrong guess is fine. A missing guess is the habit you are changing.
Test it for five minutes
Print the value, set a breakpoint, or change one thing. Use a timer. When it rings, you are allowed to ask.
Paste the error with your guess
"Here is the error. I think the cause is X because Y. Am I right? If not, give me a hint before the fix." The answer now corrects your thinking as well as your code.
What not to do
- Do not apply this to production incidents. When something is down, fix it the fastest safe way and learn afterwards.
- Do not loop. If you have pasted three errors in a row from the same bug, stop and read the code the AI wrote.
- Do not ban the tool. You will switch to another one, or lose a day to a missing comma.
The honest limit
Five minutes per bug adds up, and on a deadline you will skip it. That is fine. Aim for the guess on most bugs, not all. Some errors really are just noise from a tool you will never need to understand, and the AI is the right answer for those.
Questions
Is pasting errors into AI bad practice?
No. It is a normal part of the job now. The change is having your own guess first, so that each bug teaches you something about the system.
How do I know if I am improving?
Keep the guesses in one file. After a couple of weeks, count how many were right or close. A rising share is the only measure that matters here.
My guesses are always wrong. Is this pointless?
Wrong guesses are where the learning is. Each one tells you something you believed about the code that was not true. Ask the AI why your guess was wrong, not just what the fix is.