I agree. I mean, before LLMS we were also just trying to deliver whatever the prof asks of us.

I was thinking in terms of the rage I felt when someone committed and stomped over a bug I spent *hours* tracking down & testing and fixing. I think if I fix it, commit, and some viber + AI commits an entirely-new codebase (as prompt-written source is prone to be) needing that repeated, they’ll be dead in the desert by dawn.

There seems little doubt that the proliferation of this slop will result in enormous problems down the road that will be difficult and costly to clean up.
When they do, it SHOULD teach company executives to scale back if not stop the use of LLMs in their IT development shops, but, based on my experience of the lack of mgmt interest in quality improving systems analysis & design tools & methodologies back in the day, I fear it won't.

@daisy

The issues aren't things being slightly wrong in ways that the other developer might not be aware of, shit is just weird

Like a function that doesn't do any async operations returning a Task just like for fun I guess

AI/LLM doesn't understand the text it's translating, mistakes are random, no feedback, no learning on the other end + it takes more time to correct everything if it's even possible to correct it (and it's paid worse)

They want to vibe the requirements of product, vibe the design, vibe the implementation, vibe the tests, vibe the documentation, and vibe the reviews. They’ll reach a point in which they’ll need to vibe the customers, and maybe then they’ll realize how idiotic the whole thing is.

I've been trying to figure out a good way to handle this at our company. But so far it seems whatever guidelines I put in place are either fed right into their prompt or just ignored. No one seems to understand the code they're pushing off to a reviewer either.

The unpredictability of when it's useful is the big problem though: building a mental model based on non-deterministic outcomes is basically impossible.

There's an ongoing debate on whether these actually are useful or worsen the codebase in the long term but code ownership has to be non-negotiable for any of this stuff to work.

Me: "Where's the corroborating evidence? Can you link me to the documentation that supports this?"

Them: "Huh?"

Me: