Sovcombank Technologies has implemented an AI-driven approach to manage technical debt in legacy projects. The team uses a system of specialized skills and a central orchestrator, refactor, to automate refactoring tasks while maintaining system stability.
The process relies on a central AGENTS.MD file that stores project-specific technology preferences, architectural standards, and coding conventions. This file serves as the single source of truth for AI models during the refactoring process.
The team developed specific skills for recurring tasks, such as unit test generation, utility extraction, hook migration, and JSX component splitting. A dedicated skill, tailwind-to-scss-conversion, was created to standardize styling, as Tailwind was only applied to approximately 20% of the project.
The refactor skill acts as an orchestrator that analyzes components, creates a step-by-step plan, and executes necessary sub-skills. It ensures that changes are verified against existing behavior and prevents unauthorized scope expansion during the refactoring process.
To evaluate the effectiveness of this approach, the team conducted a study using a sample of 36 components. They compared manual refactoring against the AI-assisted method by measuring total refactoring time, the number of code review comments, and the frequency of post-change bugs.
