I'm sorry but anyone who says AI produces good results because "100% test coverage" doesn't understand software engineering.
Source: I have a PhD in software engineering
2 Boosts
0 Likes
the idea that vibe coding has improved code quality is simply laughable. even Karpathy (guy who coined the term vibe coding) wouldn't agree
like idek know how to engage with that claim, because it's so ridiculous, but I hope people who say that are getting paid well for saying it
like have you looked at the code LLMs produce? it's bad. and on top of that it must be bad, because the LLM was trained on bad code
the #1 lesson about tests is that they're only useful *when they catch bugs*
a test that never catches a bug is definitionally useless
the emprical science has always, long before vibe coding, *only sometimes* supported coverage correlating with quality
the actual situation seems to be that vibe code benefits more from coverage than human written code, probably by forcing the harness to iterate more
also, more production (read: non-test) code is always worse. always. you literally can't have a bug in code that doesn't exist