Ask HN: What tools are you using for human code review of AI-assisted code?
I and many of my colleagues are now generating AI-assisted code at an incredible rate, but the real quality gate for our projects is how thoroughly this code is human-reviewed to ensure it's not just correct, but architecturally sensible. AI code review tools like coderabbit, Copilot, or even pointing Claude Code at a PR are good at finding bugs and style nits, but less effective at spotting duplicate code, module cross-coupling, or bad separation of concerns. I'm finding GitHub's PR interface is becoming unmanageable for the size and noise of these reviews, especially with agent reviews and copy-pasted agent output mixed in. What tools or processes have you found that work well for streamlining human review of AI-assisted code?
We've started using a custom linter that flags architectural issues like duplicate code and cross-module dependencies, and we run it as part of CI. It doesn't catch everything, but it cuts down the noise so human reviewers can focus on the important stuff.
I've been using a diff-based review workflow with a tool that lets me collapse AI-generated hunks and only expand the ones I need to inspect. It helps manage the volume, but I still have to manually check for architectural problems.
We've switched to a two-stage review: first, an AI tool does a broad pass for bugs and style, then a senior engineer does a focused review on architecture and design. We use a checklist to guide the human review, which has helped catch issues the AI misses.