Claude cannot properly refactor its own slop

Our code base heavily rely on AI-made copy pasting. This was manager deliberate decision and they explicitly forced engineers to not think of common libs, but write MDs instead for AI-baded copy paste. And today when doing a massive common logic fix in \~70 different service I found that Claude Code hallucinates heavily. \- It simply left some places unchanged \- Some common places were refactored differently and no one knows why \- Tests were adapted to make wrongly refactored logic pass This all results in a few prod issues because the total amount of diff to he reviewed was \~30k lines. No one can review such large PRs properly