I've run into quite a few pitfalls recently using AI to write code, and I've learned a hard lesson: never rely on AI to look up critical data for you—you have to figure it out yourself first.

These AIs are like someone in the honeymoon phase of a relationship—they'll say yes to anything you say. The problem is, just to make the code run, they’ll make up data out of thin air. On the surface, there are no errors, but the solutions are so shallow they're basically useless.

So now my process is: first manually confirm core information like data sources and API parameters, then let the AI handle the specific logic. Otherwise, get ready for some rework.