Diff or Green Check Mark: Which Would You Go For?

An agent completed a feature last Friday in about four minutes; the tests were successful, and I almost merged it without checking the diff. However, something was off in my mind, so I checked it out. It silently changed the helper function used by three other screens. Nothing went wrong yet; it would, the next time somebody would do something with billing.

Now, I follow a certain sequence when reviewing diffs. First, I check the list of files that were modified since an unusual file name gives a quick hint. Then I read the changes to the shared code only.

Also, I have started asking the agent why it did something out of the scope of the task. The responses are quite informative. There can be a perfectly valid explanation. Or it can boil down to the fact that it just looked cleaner.

Speed is the whole point of vibe coding, and I do not want to sacrifice it. However, a green checkmark just means that my tests were satisfied; it does not mean that it was a correct decision to make.

How much of the diff do you really read? Do you have any guidelines for slowing down?