I disagree. If the code is auto-accepted, you’ll find that can’t check it before it become your code. There needs to be a way to distinguish between suggested code and actual code.
I don’t think that there’s a dispute about the role (and value) of reviewing code. However, there is a userbase of cursor (I’m part of that) that wouldn’t have opened an IDE and write some code. I hear of many Product Managers who use Cursor to build proof-of-concepts, prototypes, or even pretotypes using it.
I find myself accepting all changes, but often questioning the approach Cursor takes in building or troubleshooting.
This is the way. Most developers use Git to track changes. The Git view I always use has two panels, with the highlighting old code in red, and the right highlighting new code in green.
After the AI makes changes, the left still has old in red, but now the right has both old (in red) and new(in green). It’s just a jumbled mess that’s hard to read.
So I’m constantly hitting “accept changes” just to get to a nice Git view to review before committing. The extra step is annoying. Let it change whatever it wants, and let Git do its thing. It’s bizarre to have this change tracking double-implemented.