Can I Trust the Code AI Writes?

↗

AI helped me build several projects I had wanted to make but kept putting off. Being able to try an idea by making a first version has made a real difference to me.

I still don’t fully trust it. It can deliver code and a thorough explanation, and the work looks finished. But sounding convincing doesn’t mean the program does the right thing.

It says it’s done. I still need to try it.

Take a save button. A success message appears when I click it, but is my work still there when I reopen the app? The program starting and the button responding don’t answer that question. I need to try what comes after saving too.

↗

An illustration of the hypothetical save-button example, not a product screenshot or a recorded incident.

Having AI write tests helps, but if it misunderstands the requirement, the code and tests may follow the same mistake. They agree and the tests pass. I still need to check the result I originally wanted, rather than just count how many tests it says passed.

So I look at what it actually tried, what happened when the feature was used, and how it behaved when something went wrong. Asking it to explain again why everything works still feels like leaving a step out.

When code is wrong, I can ask AI to keep fixing it. But if all I do is forward error messages, it becomes hard to tell whether a change solves the problem or gets around it. The error may be gone for now; when I want something different later, I still need to understand what those fixes left behind.

I ask it to explain the parts I don’t understand, and look things up myself too. For now, I at least want to know whether a change brings the result closer to what I need.

Will checking it always need me?

Clarifying requirements and checking results still need my attention today. I don’t want to take that as proof that people will always be irreplaceable. AI has already helped me finish projects I hadn’t made before. Calling it a tool for repetitive work alone doesn’t fit that experience.

If business knowledge and judgment can gradually be handed over too, what advantage will programmers have left? I haven’t worked that out. Naming taste, creativity, or communication doesn’t guarantee that any of them will be safe either.

I’ll keep using AI. I need to check its results now; I’m not ready to say whether I’ll eventually be able to hand that over too.