You named the bottleneck exactly. It’s at the center of the thesis behind our app. The trap lies in thinking that code should be the primary level of abstraction for review.
The problem with having AI help you in the review step - for example, by using CodeRabbit - it sees bugs and issues everywhere. And it will give you so many “suggestions”, that it turns into noise. Hence the reason we don’t use AI review agents.
We realized that running those checks before the code makes it to PR is better. As a team we’ve moved to QA tooling that “fixes” your code prior to opening the PR.
Try it: have better linting rules, improve your instruction.md files, create ruff scripts, run code quality checks… and hook them to your pre-commit.
You named the bottleneck exactly. It’s at the center of the thesis behind our app. The trap lies in thinking that code should be the primary level of abstraction for review.
I’d love your thoughts -- https://thegyre.io/thesis
The problem with having AI help you in the review step - for example, by using CodeRabbit - it sees bugs and issues everywhere. And it will give you so many “suggestions”, that it turns into noise. Hence the reason we don’t use AI review agents.
We realized that running those checks before the code makes it to PR is better. As a team we’ve moved to QA tooling that “fixes” your code prior to opening the PR.
Try it: have better linting rules, improve your instruction.md files, create ruff scripts, run code quality checks… and hook them to your pre-commit.