The history of software engineering tools has been oriented toward a singular direction: translating human intent into machine-executable form. The emergence of AI-powered coding agents capable of autonomously generating production-grade backend infrastructure code has inverted this direction. As AI systems now author between 30% and 50% of enterprise code, the critical challenge is no longer how humans instruct machines, but how humans comprehend what machines have decided to build. We address this gap through two contributions. First, we conduct a systematic literature review (SLR) following PRISMA guidelines, synthesizing research across three domains — human oversight of AI-generated code, cloud infrastructure visualization, and human-in-the-loop software development frameworks — drawing from ACM Digital Library, IEEE Xplore, and Semantic Scholar (2020–2024). Second, we propose the Visual Infrastructure Comprehension Framework (VICF), comprising five design principles, a four-layer Visualization Model, and a five-state Interaction Model.
MyungWoon Oh (2026) studied this question.