[al_borland]

Some others must have a very different relationship and experience with Copilot than I’ve had.

Any time I ask Copilot to do anything mildly complex, the results aren’t very good. I find it works best at the line or function level. If/when it gets it wrong and requires debugging, it hasn’t been able to debug itself. This means I need to understand what it’s doing. I don’t know about anyone else, but I’d much rather debug my own code than someone else’s, and they includes AI code.

Maybe I’m not being precise enough in what I’m asking for. However, I generally learn the details and points where I need to be precise through the act of writing the code, testing, and making design decisions along the way. Without that process, I don’t know how I can step in knowing nothing, precisely ask Copilot what I want, and get good code that just runs. It’s not magic.

So far my best Copilot experience was asking it to replace a bunch of jQuery with vanilla JavaScript, which it seemed to do quite well. This didn’t expand my horizons or creativity, it just saved me a bit of self-imposed toil, as no one was actually asking me to do this.

[PaulHoule]

It competes with web search engines looking up answers for how do do little things like "How do I get the length of a string in PHP?"

It gives digested answers which are simpler to process in most cases but it is slower than a web search engine so it saves more effort than time. For CSS it is pretty good because with CSS you have the choice of looking up authoritative answers in the 50 or so standards docs (+time +effort) as opposed to taking your chances with community answers on the web, Copilot comes close to "read the manual" in accuracy but with (-time -effort)

[JohnFen]

You aren't alone. I, too, haven't found copilot to be all that useful.