Bankdata modernizes mainframe by automating COBOL migration

Bankdata, a consortium of Danish banks, faced the challenge of modernizing their vast COBOL-based mainframe legacy systems to cloud-native platforms due to growing technical debt, rising maintenance costs, and limited access to legacy experts. The organization aimed to retain more control over project costs and intellectual property, moving away from traditional approaches heavily reliant on global system integrators. Leveraging state-of-the-art Microsoft technologies, Bankdata and partners developed a modular, agent-based migration factory that uses multiple orchestrated AI agents to analyze, convert, and test COBOL code into maintainable Java running on modern platforms. This system underpinned the transition by extracting business logic, visualizing and mapping dependencies, and ensuring that legacy business processes are accurately transformed. Sophisticated orchestration with Microsoft Semantic Kernel enabled precise management of worker agents, intelligent handling of code context, and conversion consistency. Using GPT-4, GitHub Copilot, and Azure OpenAI, the framework delivers robust code translation, dependency mapping, call chain analysis, and quality assurance through test suite automation. The project significantly reduced manual workload, improved code quality and maintainability, and accelerated transformation timelines, all managed in-house at Bankdata.