Recently, I’ve started asking a question in code reviews#
that would have sounded absurd two years ago:
Which part of your code could you not fix
if it broke tomorrow and AI were unavailable?
If the answer takes a while, I know where we need to start.
Being able to explain AI-generated code is not enough.
If you own it, you need to be able to fix it.
For me, that means three things:
- being able to explain why the code was built this way
- knowing what AI must not be allowed to decide in this system
- recognizing when its suggestion should be rejected
That is precisely what I test at the whiteboard.
Not whether someone can rewrite every line from memory.
But could they get things working again without AI.
Which part of your code would be difficult today?
#SoftwareCraft #AI
