[userbinator]

Good luck trying to debug the code you've generated without knowing anything about what it does.

[denkmoon]

Is this really any different from debugging code that has been built by hundreds of people over 10s of years and you're just the latest schmuck working on it?

[DanHulton]

It really depends on if the 10-year-old code in question was put together by a talented team with high standards, doesn't it?

But regardless, on a large project, you're always going to have some version of this problem. At least when you yourself are getting in there and writing and understand the code, you have a half-decent shot at debugging in a reasonable period of time. When even you are vibe coding your additions to the pile, you're all the way back at square one when shit hits the fan, trying to learn it all from scratch.

[userbinator]

Yes, because you actually had an incentive to learn about the code.

[plorkyeran]

Maintaining legacy code built by a huge team over decades is sort of famously difficult and the productivity expectations on those teams is incredibly low. A project getting into a similar state after only months would be disastrously bad.

[ofjcihen]

Yes. Because you know the language

[waltbosz]

This is true. I've been using LLMs for code generation, and it does stuff I don't understand well. Also it makes mistakes, bugs.

What I do find helpful is use it to ask for suggestions for how to achieve some result, then learn from the code it gives me.

But even then, going back to this code later is difficult because I didn't go through that period of the failure research success learning loop.

[mcny]

Maybe at the high end of the talent but most of us are already there. I wouldn't admit this to my manager but a lot of the time if you made me explain what actually happens at the machine level, I have no idea.

[waltbosz]

I think there is an acceptable level of black boxedness in our profession.