Currently, it's basically entry level for anything mildly complicated and it can't troubleshoot itself, so it needs to be coached into giving the right answer. So it's great for boilerplate and API lookups but if you were using a typed language before and had a decent IDE, this was already a solved problem. It seems impressive because it's very fast but that advantage goes away when you then have to be like, "hmm, well that's not quite right" and then you coach it into being correct (or close enough that you stop).
If your team can't autocomplete APIs, and if you can't run everything locally so you can step through your code with a debugger, and if you don't have good tests so you need an AI to tell when you broke something (see also profiling, telemetry, release quickly and confidently, etc) - you're going to get a million percent more mileage just doing these sort of basics correctly (and fast) vs investing in AI because the open secret is that most of engineering time is spent reading and debugging code rather than writing it. This should be table stakes but so many engineering teams (even at bigtech companies) can't do the above. So they're all signing up for copilot but they still iterate with logging statements and catch regressions in production.
If your team can't autocomplete APIs, and if you can't run everything locally so you can step through your code with a debugger, and if you don't have good tests so you need an AI to tell when you broke something (see also profiling, telemetry, release quickly and confidently, etc) - you're going to get a million percent more mileage just doing these sort of basics correctly (and fast) vs investing in AI because the open secret is that most of engineering time is spent reading and debugging code rather than writing it. This should be table stakes but so many engineering teams (even at bigtech companies) can't do the above. So they're all signing up for copilot but they still iterate with logging statements and catch regressions in production.