@andreitr.eth
The Megapot eng team is shipping a lot of code. We’ve automated code reviews with Codex and Cursor’s BugBot, and we enforce a strict no-warnings ESLint policy. That said, human reviews are still essential for catching architectural issues and that’s now our biggest bottleneck. Our goal is to merge PRs within a few hours (max) of opening.
To push further, we’re experimenting with shared AI skills (Claude skills / agent skills) for common workflows like database work, queries, and UI.
The idea is to enforce architectural standards and “soft” conventions at development time, not during review.
Curious how others are using AI to enforce soft conventions, the unwritten rules, architectural intent, and team norms that usually only show up in review comments.
Are you doing this today? If so, what’s working (or not)?