We spent months working on this and certainly didn't vibe code it. We do use LLMs but primarily for code review and boilerplate. We're now including a lot more tests partly thanks to it being less of a pain.
Our quality standards are too high for current frontier LLMs to generate much we could use directly. On the other hand, their code review output is extremely useful. It can find many issues we wouldn't catch even with multiple rounds of human review. It's helping us a lot with catching issues we've repeatedly overlooked.
alt Hacker News
We spent months working on this and certainly didn't vibe code it. We do use LLMs but primarily for code review and boilerplate. We're now including a lot more tests partly thanks to it being less of a pain.
Our quality standards are too high for current frontier LLMs to generate much we could use directly. On the other hand, their code review output is extremely useful. It can find many issues we wouldn't catch even with multiple rounds of human review. It's helping us a lot with catching issues we've repeatedly overlooked.