We are QA Engineers now

Writing tests at all levels of integration was always a major part of the job. As with everything else, some individuals and organizations have done it better.

These are four great general guidelines to have in mind when approaching vibeslop codebases. Thanks!

I have been finding that enforcing generic good code practices can help LLMs trim through the previously done slop, not only "how can I test this?" but also "is it commented? do the comments explain why things are what they are?", "are functions line size below a given threshold? where are the biggest offenders?" or even cleanear stuff, like where are the side-effects? are all components pure? can we decompose a given output to the set of state transformations that introduced it?

I've been struggling with making LLMs produce defensive code in these vibeslop landscapes that we have to deal with nowadays.

AI writes the code, but its me who has to sign off on it (for bad or for worse) and manage expectations.

Always have been! It's just been greatly deprioritized. LLMS really closely mimic humans - I know the whole reason I write tests is because I am severely imperfect and I like the feedback of a verification check.