In software development, we have high hopes for generative AI, but we run into limitations with legacy systems because language models—just like humans—struggle with tangled codebases. The key lies in making legacy systems consumable by AI—for example, through methods such as “Domain-Driven Transformation,” which breaks the code down into manageable, domain-structured units. This article explains why this is necessary, how it works, and reports on concrete experiments showing where AI can already help with this transformation today.