It's so upsetting to see people take the powerful tool that is an LLM and pretend like it's a solution for everything. It's not. They're awesome at a lot of things, but they need a user that has context and knowledge to know when to apply or direct it in a different way.
The amount of absolutely shit LLM code I've reviewed at work is so sad, especially because I know the LLM could've written much better code if the prompter did a better job. The user needs to know when the solution is viable for an LLM to do or not, and a user will often need to make some manual changes anyway. When we pretend an LLM can do it all, it creates slop.
I just had a coworker a few weeks ago produce a simple function that wrapped a DB query in a function (normal so far), but wrote 250 lines of tests for it. All the code was clearly LLM generated (the comments explaining the most mundane of code was the biggest give away). The tests tested nothing. It mocked the ORM and then tested the return of the mock. We were testing that the mocking framework worked? I told him that I don't think the tests added much value since the function was so simple and that we could remove them. He said he thought they provided value, with no explanation, and merged the code.
Now fast forward to the other day and I run into the rest of the code again and now it's sinking in how bad the other LLM code was. Not that it's wrong, but it's poorly designed and full of bloat.
I have no issue with the LLM - they can do some incredible things and they're a powerful tool in the tool belt, but they are to be used in conjunction with a human that knows what they're doing (at least in the context of programming).
Kind of a rant, but I absolutely see a future where some code bases are well maintained and properly built, while others have tacked on years of vibe-coded trash that now only an LLM can even understand. And the thing that will decide which direction a code base goes in will be the engineers involved.