99.9% of the code in my projects is written by the Agent. And I make it keep test coverage at least 75%, so it can catch its own bugs. Pretty often (if not always), when I move away from that rule, I end up with either new bugs or regressions.
But suddenly I started wondering: are the tests really better in quality than the code itself?
Overall, with high test coverage — which the Agent also writes — I get a pretty solid product. But what do you think about that?