Self-Repairing Codebases: Agentic AI for Continuous Legacy Migration

ABSTRACT

The migration of legacy codebases (e.g., COBOL, Fortran) to modern architectures is a notoriously slow, error-prone, and expensive human endeavor. We introduce 'AutoMigrate,' a persistent, multi-agent AI system designed to operate continuously in the background of enterprise repositories. Utilizing an iterative 'Translate-Test-Refactor' loop, the agent swarm autonomously isolates legacy functions, translates them to Rust, generates comprehensive unit tests, and submits pull requests for human review. Over a six-month deployment at a major financial institution, AutoMigrate successfully modernized 1.2 million lines of legacy code with zero production downtime.