Comment by altern8
1 month ago
I see AI as an awesome technology, but also a like programming roulette.
It could go and do the task perfectly as instructed, or it could do something completely different that you haven't asked for and destroy everything in its path in the process.
I personally found that if you don't give it write access to anything that you can't easily restore and you review and commit code often it saves me a lot of time. It also makes the whole process more enjoyable, since it takes care of a lot of boilerplate for me.
It's definitely NOT intelligent, it's more like a glorified autocomplete but it CAN save a huge amount of time if used correctly.
The safety practices you describe are basically the right mental model: assume it's fallible, keep writes reversible, review everything, commit often
Yes, it's been working well.