I used AI to investigate a large open-source Windows codebase. The work included building the software from source, locating relevant behavior across unfamiliar files, checking live process and module information, tracing configuration defaults, and translating the findings into clear operating guidance.
The value is not simply generating code. It is entering an unfamiliar system, forming testable hypotheses, inspecting the evidence, isolating the likely cause, and explaining it so someone can act on it. That is the same root-cause discipline I have used in analytics since my consulting days.