This is what happens when you give a new tool to people without educating them on how to use it properly. Problem with AI is that it’s so new to us all it’s difficult to find an authority on the subject.

In my experience it’s useless for fixing bugs or identifying flaws (mostly). That’s still our job, ensuring the code is generated as we intended it to be.

AI tools are most effective when you already know what you want to write, or if you need help finding alternatives to a poor solution you’ve implemented.

This is the new argument from the AI unbelievers “it’s an illusion, it’s imagined, it’s not real, it’s actually SLOWER”.

Then they demand proof that AI is faster which is of course very difficult to quantify, to which the unbelievers say HA! See! You can’t prove it so it’s all fake!

You can't get something from nothing. You have to apply an equal amount of context for whatever you want the output to do. Maybe you'll need to be specific enough that you might as well have just done it yourself.

If you don't do that, you will get A Thing, which may may not be what you want, which may or may not do what you want.

Describing what you want is non-trivial effort and a skill. Unfortunately, the Read My Mind button is still a long way off.

When you cede your brain to the LLM, you can go into extended zombie loops where the LLM keep pushing crap to you, you keep correcting it, until it goes back to where it started.

maybe, the devs who struggle most and are slowest are also the ones most likely to use the llms?