Application modernization has always been hard, especially for complex, hardware bound and proprietary systems. Agentic AI gives us a new chance to make these projects technically and economically viable. But producing modern code does not prove successful modernization. This talk shows how to assess a candidate, build the right context, work in controlled increments, and combine human, AI and deterministic verification to define what good looks like and prove the result actually works.
