Throw a Prompt at your IDE and see it get done!

Even as a heavy user of agentic IDEs—and someone building frameworks for GenAI orchestration myself—I’m a bit torn.

On one hand, these tools are amazing. You can almost treat your IDE like a black box: throw a prompt at it, judge the application behavior and test results, and let the model do its thing until it works.

On the other hand, this only works if you understand the system extremely well.

Because someone still has to understand all the edge cases, side effects, framework gotchas, and hidden requirements in order to properly assess whether the code is actually done.

The repeating pattern I observe when I do what people call “vibecoding” looks something like this:

prompt → it works  
try it → it breaks  

prompt again → it works  
try it differently → it breaks  

you run out of time / deadline → ship it  
production → it breaks again

Sound familiar?

And yeah… this is literally how a junior developer programs.

Is this a bad thing?

I think... No.

In fact, leaning into this workflow has made me realize something important: Vibecoding turns even the most Senior Developers into Junior Developers again.

And honestly? That might be a good thing.

But at its core, I think "Vibecoding" has surfaced a question we maybe never really answered — even before LLMs existed:

What the hell is software engineering actually about?