Mixture of Experts Mainframe modernization explained: COBOL and AI
24 snips
Feb 27, 2026 
Kaoutar El Maghraoui, AI systems researcher focused on model optimization. Maryam Ashoori, AI governance lead specializing in accountability and evaluation. Skyla Loomis, mainframe modernization leader integrating modern tooling and AI. They discuss AI-powered modernization of legacy systems, marrying LLMs with static analysis for COBOL, hybrid small/large model architectures, infrastructure and adoption barriers, and security and ops for agentic AI.
AI Snips
Chapters
Transcript
Episode notes
Mainframe Modernization Is Integration Not Migration
- Mainframe modernization is not simply migrating away from COBOL or changing languages.
- Skyla Loomis explains mainframes are a fit-for-purpose, fully engineered stack optimized for specific workloads and require integration, not wholesale replacement.
COBOL Stays Because Business Logic Outlives Developers
- Enterprises want to keep COBOL because business logic and longevity matter more than language novelty.
- Loomis notes vast, decades-old codebases contain undocumented rules written by people no longer at the company, so understanding them matters more than rewriting.
Combine LLMs With Static Analysis And Humans
- Don’t rely on LLMs alone for huge legacy codebases; combine LLMs with deep static analysis and human oversight.
- Loomis recommends incremental, bite-size transformations aided by static program analysis to reduce hallucinations and validate behavior.