My AI programming agent is awesome with codes but absolutely horrible with judgment.

After three weeks of using the agent for my back-end work, the one thing that keeps coming up is not bugs but confidence. The same way it would properly refactor a function, it would also just as confidently remove a validation check, as "it looked redundant," when in reality, it was there because of a bug that I had fixed in March.

I now ask it to justify the reason for something before removing or changing it, not just its functionality. And this one action alone has greatly reduced the number of times I have to undo things.

It is curious how few people consider the question of "Why is this here?" as necessary before giving an agent free reign to alter legacy code.