AI Is Writing Code It Doesn’t Understand (JetBrains Warning)

This is a summary of a video I published on YouTube. You can click on the link below to watch the full video.

The biggest mistake developers are making with AI right now

A lot of developers think AI means they can skip thinking. That’s the mistake. And it’s going to cost them time, money, and a lot of frustration down the road.

What I’m seeing more and more is people letting AI generate entire chunks of applications without any real structure or oversight. It feels productive at first. Code appears quickly, features seem to work, and everything looks fine—until it doesn’t.

Then things start breaking. And nobody knows why. Because nobody really understands what was built.

The workflow that actually works

If you’ve been around development long enough, you start to recognize patterns that hold up. This is one of them. AI works best when you put it in the right role.

Here’s the simple loop that makes sense:

• The human defines the system — architecture, data flow, responsibilities
• AI handles implementation — code generation, boilerplate, components
• The human reviews and refines — ensuring clarity, correctness, and fit

That’s it. Nothing fancy. But if you skip the first or last step, things fall apart fast.

AI is not there to think for you. It’s there to execute within boundaries you define.

What happens when you skip the thinking

I call it “AI slop.” You get code that technically works, but it’s messy, fragile, and hard to reason about. As soon as you try to extend it, things start to crack.

This goes directly against what good developers aim for: simple, understandable systems. If you can’t explain how your code works, you don’t control it. And if you don’t control it, you can’t maintain it.

I’ve already seen teams run into this. Even companies building developer tools are warning about it. This isn’t theoretical.

Why fundamentals matter more now, not less

There’s this idea floating around that AI makes fundamentals less important. It’s the opposite.

If you don’t understand architecture, state, and how systems fit together, you can’t guide the AI properly. You also can’t evaluate whether what it produces is any good.

So you end up accepting bad output because you don’t know better. That’s where the real risk is.

A simple example that says it all

I tried having AI generate a page for a site from scratch. It struggled. The output wasn’t quite right, even after multiple attempts.

Then I changed approach. I copied an existing page, gave the AI a specific task inside that structure, and suddenly it worked perfectly.

Same AI. Different context.

The difference was structure and constraints. That’s your job.

The real takeaway

AI is like an IDE on steroids. It can speed you up, but it doesn’t replace judgment.

If you focus on architecture and clear thinking, AI becomes a serious advantage. If you try to skip that part, you’re just building problems faster.

Don’t hand over control. Use AI to execute your decisions, not replace them.

Watch the video on YouTube here 👉 AI Is Writing Code It Doesn’t Understand (JetBrains Warning)

Thanks for reading!
Stef