Retired dev here, I'm curious about the nature of "the mess". Is it buggy AI-generated code that got into production? I know an active dev who uses ChatGTP every day, says it saves him a hell of a lot of work. What he does sounds like "vibe coding". If you're using AI for grunt work and keep a human is in the workflow to verify the code, I don't see how it would differ from junior devs working under a senior. Have some companies been using poorly managed all-AI tools or what? Sorry for the long question.

I imagine you aren't talking about large companies that just let ai loose in their code base. Are these like companies that fired half their staff and realized llms couldn't make up for the difference, or small companies that tried to make new apps without a proper team and came up short?

Same thing happened during the outsourcing craze of early 2000s. Everything and I mean everything moved to India or Philippines. There's even a movie about it because it was so common. I and everyone else lost our jobs. about a year later the contracts expired and we all got jobs back and outsourcing is used in balance. Eventually ai use will be balanced I hope. It cannot replace us. Not yet anyways.

essentially, from what I've been dealing with, most if it is their offshore people using the AI to completely do the job from start to finish and no one is verifying anything. So it's not even vibe coding, it's "here's a prompt, build it, i'm pushing it to production" coding.

To be fair, 0.2 + 0.1 = 0.30000000000000004

AI as it exists today is only effective if used sparingly and cautiously by someone with domain knowledge who can identify the tasks (usually menial ones) that don't need a human touch.

AI as it exists today is only effective if used sparingly and cautiously by someone with domain knowledge who can identify the tasks (usually menial ones) that don't need a human touch.

The biggest point is that you must be an expert in the field you are using it in. I rarely get fooled by hallucinations and stupid bugs because they are glaringly obvious to me. The best use case is having the llm write code for using a library that has poor documentation, that am going to use once, and I am too lazy to learn.
These tools are scary when used by juniors, they are creating more work for everyone by using llms to code. I just imagine myself using this when I was a fresh grad, it is terrifying. It would have only been one step up from vibe coding.

This is the best description of current AI I've seen so far.

The biggest point is that you must be an expert in the field you are using it in. I rarely get fooled by hallucinations and stupid bugs because they are glaringly obvious to me. The best use case is having the llm write code for using a library that has poor documentation, that am going to use once, and I am too lazy to learn.
These tools are scary when used by juniors, they are creating more work for everyone by using llms to code. I just imagine myself using this when I was a fresh grad, it is terrifying. It would have only been one step up from vibe coding.

We've hired a bunch of Indian guys who are using AI to do their work... the results are marginally better than either approach independently.

it doesn't look like just a gimmick anymore.

It still does