LLMs vs. brownfield reality: Why refactoring enterprise systems is hard

This article argues that while Large Language Models (LLMs) hold promise for modernizing legacy systems, they struggle with the complexities of real-world enterprise applications. LLMs lack the necessary domain knowledge, runtime awareness, and ability to handle large codebases – leading to issues like inconsistent outputs and potential production failures. The key to successful modernization lies in using LLMs as execution engines for pre-defined, data-driven refactoring plans, rather than relying on them to design the architecture itself, and combining them with runtime analysis and human oversight.