Dev Tools · 1h ago
AI agents and AST filtering cut i18n leaks from 300 to 60
A team used four parallel AI agents to detect hardcoded Japanese text in a bilingual app, reducing 300 candidates to 60 real leaks. AST-based filtering eliminated false positives from translation tables and language branches. The cleanup revealed that English users had been receiving Japanese emails from Stripe for months.
Meridian48 take
The approach is clever but the 80% false-positive rate shows AI still needs heavy mechanical filtering to be practical for code analysis.
Read the full reporting
Sweeping i18n leaks with four parallel AI agents — from 300 candidates down to 60 real bugs →
DEV Community
ai-agentsinternationalization