I’ll tell you how I do it. That doesn’t mean it’s the best way or anything like that. First, you need to have plenty of unit tests (which the AI can also help with); that ensures that even if the AI makes changes, it has to do so under strict rules. Then, it’s also best to have a technical specifications document where you outline what you want to do and under what conditions.
From there, the next step is to run a thorough set of running tests on the results generated by the AI (because it’s faster than reviewing the code itself). Then, once you have something consistent, you can start tweaking the code yourself.
Ultimately, what I always do is let the AI write the commit message. It’s infinitely better than me summarizing the change, and it’s even easier for me to read.
Yes. The first thing I did was create an extremely detailed specification document where I wrote down what I wanted, how I wanted it and where I wanted it. For me, that’s the most important foundation. With that, you’ve already done a lot of the work. In fact, I even go so far as to write pseudocode in some cases.
From there, you use that as your ‘bible’ and stick to the script you’ve written for yourself. Obviously, as you go along, unforeseen issues may arise, and in that case, make a note in the spec, find the best approach, and carry on.