One thing I’ve noticed with AI coding tools is that the issues aren’t usually syntax errors they tend to appear at the system level ,the generated code often looks correct locally but can break integration points between services or APIs ,do you guys here have seen similar patterns when reviewing AI-generated PRs