oscarestudillom

something that works well to me is trying to make very compact PRs, with just one intention, and not divagate into multiple ideas at the same time.

also at my company we use Revix AI for reviews which do not remove the human but gets most of the little things and allows us to keep proper code in a team where most of juniors and mids use AI without much control

Nah, I don't think we are at this point yet. AI without control just creates non usable products. We don't literally write the code but we still give direction and review output.

We just found the bottleneck moved to the reviewing part, and AI can help there too, we tried many AI reviewers and even mostly just add more noise and slop we found some that are quite good and even do not replace humans they take most of the consuming work out.

Once we get to the point code is not the problem, remember that we started coding because we wanted to create software that solved problems. What I mean by this is don't forget we are here to solve problems, not to write code.

We keep human review because AI coding is not yet good enough. We use though an AI reviewer first, before the human even see the PR, that helps us filtering most of the bad code details so human don't lose time and focus on important things. Be careful though when choosing an AI reviewer, and never use claude/codex directly for this, they will just add noisy comments and instead of helping humans will just add more work to filter which are good and which are just noise.

What worked best for me is to start doing things by hand, trying to do the architecture and the standards I really want. Then, AI works really well on following my conventions. Therefore it gets easier if I want to check codebase, because it's already organized as I initially planned, even all the new code is not anymore written by me directly. Then I also use Revix AI as a reviewer on my PRs that really helps me filtering anything that gets out of the standards and conventions I like.

Other than that, on other projects that I fully vibecoded, it's just impossible to keep with the codebase. Just blind trust the AI models on that, and never do it on a serious production product.

Evidence is how precise the findings are, and how often people act on them. More comments are not evidence.

Volume without action is just noise.

We tried multiple AI Code reviewers and LLM based review bots and they are all the same, a lot of volume with no real action after that.

Then decided to build Revix AI, where the main focus is basically to revert that. Prefer less comments but act on all of them even if we miss some things rather than get 20 comments that I will end up not reading and in the end get more issues into prod.

Agree that shared understanding is the hard part. Bugbots find bugs. Review is how the team stays aligned on the change. Visuals help a lot for that first pass.

Another cause of burnout for us was noisy bot comments on top of diffs that already look AI-written. We only want findings we would actually act on.

I checked your skill and looks really good, however I find it to be very complex already, we do many PRs and maybe visualizing graphs for every one is too much. I think ideally could filter the complexity of PRs and adapt the visualizer depending on that, being just two sentences on some and complete graphics on some.

I myself created Revix AI and we internally created some rules that we found that served educational purpose as well, we started linking that rules to the PR comments that Revix did so the devs could access that information if they wanted. Following what I said, it's always better to keep it simple and only expand if the human wants to get deeper or if the pull request requires it.

we are using Revix AI, works really good on repos that already have some standards and patterns from good devs. the reviewer catches most of the things so when the senior reviews he just have to focus on more elevated things like architecture, etc.

moreover, try to enforce having AGENTS.md files on your repos, and rules created by the senior devs specific to your repo, its not perfect but also helps quite a lot

At my team we added AI review because seniors were tired of always commenting on the same things, mostly because not everyone uses AI properly to produce production ready code. We tried coderabbit, qodo, and they all produce an amount of noise that end up being useless. A colleague and I then decided to create an internal solution that we ended up publishing under Revix AI. We have multiple control systems to avoid noise and only comment real issues, and we prioritize standards existing in the repo over llms knowledge. This one works so good for us and I don’t understand why the others don’t work like that.