Well Ive ad a great time using LLMs to sandbox a dozen implementations and then investigate the shortcoming and advantages of different implementations.

Mistakes happen a lot but they can be managed on a small MWE with a couple of tests.

It’s how the tool is used more than any given tool being bad.

I understand your point and you’re not wrong. However, I’m not wrong either and you should take a second look at how you might use these tools in a way that makes your life easier and addresses the valid limitations you’ve described.