Hacker News
Tell HN: Fewer PRs done with proper prompting, review, and refinement wins
*3 thoughtful PRs a day > 40 poorly thought ones no matter how many AI agents reviewed them.*
4lx87
[-]
Competent engineers committing working increments of software straight to main and continuously integrated and deployed > PRs of any sort.
kypro
|root
|parent
|previous
[-]
I did this on a small microservice for a project recently where risk was constrained and I didn't see much value or appetite for human reviews. AI review, if clear, merge straight to main.
I suppose the risk is you have a vibe coder with AI psychosis blasting stuff into main on a mature project? Do you have have any sense for level of seniority, team size and size of code base that would allow this?
I get the sense everyone would need to be very senior 10+ years experience, with a relatively small team (< 10), on a medium sized and relatively modern code base?
4lx87
|root
|parent
[-]
PRs don’t actually solve the “but what if someone ships crap” problem. Taking capacity from competent engineers to review and correct other employee’s code defeats the purpose of employing people: to add capacity.
Juniors/employees in training should be shadowing or paired with someone who is competent until they are competent, and assigned work at their level.