Sweeping i18n leaks with four parallel AI agents — from 300 candidates down to 60 real bugs

Short summary

A development team used four parallel AI agents to systematically hunt for hardcoded Japanese text across a multilingual codebase, reducing over 300 candidates down to 60 real bugs. AST-based false-positive filtering automatically removed hits from translation tables and already-branched code. The investigation uncovered a critical bug where Stripe webhook emails were sending purchase confirmations in Japanese to English-paying users, plus principles for consolidating language-detection logic.

• •Parallel AI agents reduced 300+ i18n hardcode candidates to 60 real bugs in a multilingual codebase
• •AST-based filtering eliminated false positives from translation tables and already-branched code
• •Key bug found: Stripe webhooks sending Japanese emails to English customers; solution demonstrates language-branching consolidation pattern

Generated with AI, which can make mistakes.