Miki Tebeka retweetledi

There's a take going around AI coding circles that sounds really appealing.
"Tests are 100%, so who cares about code quality? Just regenerate."
Two things bother me about it.
First, if your test suite had 100% coverage of the bug you just found... it wouldn't be a bug. It would be a failing test. The coverage didn't protect you, so why is it the safety net for regeneration?
Second, there are multiple valid solutions to any problem. Regenerate the software and you might get something that passes every test but behaves slightly differently. Different UI. Different output format. Different error messages.
It might all be correct, but users only care about their experience.
Code quality matters. Tests can't encode everything your users expect. Once the code is in production, you'll have to support it, not regenerate it.
What’s the alternative? Specifying every single behaviour, including all the UX, in so much detail that you'll struggle to ever ship anything.
English

















