I've have played around a lot with ChatGPT, Claude, and local models. I primarily use my own local models to assist in coding, and even then I rarely use it.

What I have found, for small tasks, it can work really well. Large tasks are almost always going to be more work to troubleshoot than just write it yourself.

It is great for brain storming ideas on how to solve a problem you are having difficulty with.

It won't replace developers, that's for sure.

Another thing to mention, it can introduce massive security problems using old libraries and even suggesting compromised packages.