Mainframes remain the hidden backbone of the Department's financial, logistics, and contract management operations. Systems written decades ago in COBOL and Assembler still execute workloads the mission cannot pause, and they do so with a reliability that modern platforms rarely match. Yet these systems now face a convergence of pressures: the specialized workforce that maintains them is retiring faster than it can be replaced, documentation has drifted far from the running code, and forty years of accumulated technical debt make every change slower and riskier. Full rip-and-replace modernization has a long, well-documented record of cost overruns and outright failure, but standing still carries its own risk. The question facing IT and program leaders is no longer whether to modernize, but how to do it without disrupting the mission.

This session presents lessons learned from in-flight modernizations. It details a phased, automation-first methodology, delivered with DDC's GENOME (Generative Optimization and Modernization Engine), built for high-stakes environments: using large language models to translate legacy COBOL and Assembler into modular, maintainable services; recovering business logic that exists only in the code itself through graph analysis and model-assisted reverse engineering; re-orchestrating decades-old batch processing into modern, observable workflows; and migrating hierarchical data into relational models. The focus throughout is on the guardrails that make AI-assisted modernization defensible: test-driven validation that treats existing system behavior as the specification, human review at the decisions that matter, and traceability that preserves original business rules for auditors and accreditors.

Equally valuable are the lessons learned. The session is candid about where automated translation accelerates the work dramatically, where it has to be constrained, and where human expertise is irreplaceable, along with the practical realities of running this kind of program inside federal security and accreditation boundaries.

The session includes a live demonstration of GENOME, the engine behind this modernization, showing how legacy code is analyzed, converted, and validated in practice.