Comment Re:More a reflection on the state of the tools (Score 1) 50
And sometimes that works.
It's conceivable the issue here is one of habits and training - i.e. that one day we'll use such tools all the time, and have developed the right kind of instincts to code-review the suggestions that come up. Having a good feel for where tools are reliable, and where they need special attention is nothing new - and perhaps that just a question of practice here too.
Mind you, it might not be. It's sometimes harder to read code than to write it, so even if AI could write code just as good as the programmer, it's not obvious whether it's actually more efficient or in other ways better to merely select and review the AI-generated code vs. having a mental model and having a human write it. It may well be the reverse, at least for a while - that it's best to have a human use simple and predictable tools, and have an AI review the code afterwards to check for stupid gotchas. Or maybe AI needs to get a lot better before it's valuable for coding.
In practice I'm pretty unimpressed so far. It's really cool to see it work, but it's so unbelievably unreliable, that it sure comes across as a neat tech demo rather than something really useful. Then again, maybe I'm just not finding the best places to use it...