Selfhosted

Yup. You really don't want the maga cult monitoring your cycle. If you stopped menstruating for a bit you must be pregnant. Where is the baby? Omg you murdered the baby by taking Tylenol!

Appreciate that!

Imagine you are on the ground under your car and need a different tool. You ask for it and somebody hands it to you. That person is young and inexperienced. It is up to You to check if it's the right tool, and if not pass it back (and in this example tell the person about the error and help them correct it).

And sure, You can always crawl out and get the tool yourself and sometimes that is the only option and in coding terms in my opinion best practice. But you can be faster with your helper. Use it appropriately and see how it affects your work. And that's the point, your work. Don't pass responsibility or thought off to AI.

Why?

It makes sense to try to give users an idea of how robust a project is, but the exact details of the tools involved in its creation rarely add much to that. It gets a little weird with LLMs because they allow someone with no programming skill to create software that appears to work, which ought to be disclosed; "I don't know what I'm doing and I asked a robot to make this" does indicate unreliable code. A skilled developer having an LLM fill in some extra test cases, on the other hand can only make the project more robust.

You can see that I use some of metrics, like test coverage, estimates and so on to prove its validation as potentially serious project, that will grow from a pet one.

Partially agree, but I do know how to code and use it as a tool.

No one does that in a project they’re building for themselves.

Speak for yourself, I always did that and I found it easier with LLMs nowadays.

I hate most AI shite with a passion but when it helps my colleagues write commits which are more than "add stuff", "fix some things" I'm fine with it.

I rarely use AI to generate code, usually only when I need a starting point. It's much easier to unfuck AI code than to stare blankly at a screen for an hour. I'd never commit code I don't fully understand or have read to the last byte.

I hope OP is doing the same. LLMs fail at 90% of coding tasks for me but for the other 10% (mostly writing tests, readmes, boilerplate) it's really OK for productivity.

Ethics of LLMs aside, if you use them for exactly what they're built for – being a supercharged glorified autocomplete – they're cool. As soon as you try to use them for something else like "autocompletion from zero" aka "creativity", they fail spectacularly.

Judging code quality by use of LLM in a documentation and commit messages is weird.

While I write all of my code myself and I'm against vibe coding etc., there is one place where I let a LLM write for me: readmes, commit messages and Javadoc comments.

I know how to write code but at the same time I'm shit at both my native language and even more so at English. So I let Language Models write natural language texts for me and just fix them when necessary. My documentation is more clear, grammatically correct and more detailed than in any of my previous projects, and I can focus on writing code.

And I wouldn't say "No one does that in a project they're building for themselves". I do that for projects that only I will ever see, and OP shared his project with others, so it's great that he included a clear documentation

I answered earlier, that I use AI and this is just a commit skill for an agent.

Well, not stealing, being inspired)