I was debating even answering this. You've displayed that you're perfectly okay with outsourcing your agency and thinking to something that lacks the capability to do either, and so responding seems entirely pointless. For all I know you're just a machine outputting garbage based on some script. This is a public forum viewable to third parties, so that I guess makes the effort worth it.

There's not a single problem with what you propose, but a number of compounding problems that makes the usage in general problematic, and disclosure important. For the sake of brevity, let's disregard the environmental and ethical concerns of using LLMs and supporting the companies behind them. Mind you, I think these concerns are fully valid and in a broader context should most definitely not be disregarded.

LLMs are trained on massive datasets. The companies have time and again proven their complete disregard for where they're sourcing the data. Here for example is an ongoing list of lawsuits against various "AI" companies but you obviously don't need more than eyes to see this; how else would their image models be able to reproduce images of copyright and trademark protected characters?

The ability of the LLMs to produce code means that they've gobbled up large amounts of code as well, in various languages. If these companies have stolen other types of data for use in their training, why would they not have stolen code? Code also comes in various licenses, and you can't generally mix and match these licenses freely.

Now let's paint up a scenario. Say, ten years ago, a junior developer wanted to contribute to an open source project. Maybe their commit wasn't very good to begin with, but people offered feedback, and let's say eventually the PR was approved. The person came up with their own solution to a problem, and learned a bunch in the process.

Now let's say someone tasks an "agent" with fixing a problem. The LLM churns for a bit, and then spits out a solution. Let's say it works, and it fits the submission guidelines. Let's be charitable and say that everything works out perfectly, the maintainers accept the PR and everything is hunky-doory.

Then a company comes along and notices chunks of code in the FOSS project are reproductions of what's in their proprietary, or differently licensed codebase. What happens then? Who has the responsibility? Obviously the maintainers of the FOSS project.

The sloperator in the second example didn't learn shit. They gained nothing of value, and the maintainers of the project now has to go through and remove whatever infringing code they've unknowingly added to their codebase.

A more common problem FOSS projects are facing today is a complete inundation of nonsense PRs and reports by agents run by people who aren't actually looking to improve the projects. They're just unleashing a torrent of slop, overwhelming the maintainers of the projects, and breaking existing workflows. curl for example are no longer doing bug bounties because people used agents to submit nonsense for their maintainers to wade through. curl is one of those semi-invisible projects used absolutely everywhere, and so them not being able to keep doing bug bounties is a loss to everyone.

Is prejudice a part of this? Yes. I do judge people who have elected to give up their agency and have a pattern matcher do their thinking for them. I think that's pitiable, and I'm not ashamed of admitting that. Remove my prejudice from the equation however, and the points still stands.