Sustaining Peer Review in AI-Native Academia: A Dynamic Model and Three-Layer Architecture

AI can accelerate peer review, but sustaining reviewer expertise requires more than faster reviewing. We develop a simple dynamic model linking submission growth, review throughput, and reviewer development. The model shows that, under overload, faster workflows can weaken expertise development by reducing what reviewers learn from each review. This finding motivates a three-layer architecture that combines automated reference and artifact checks, human-AI deliberation, and academic credit beyond conference acceptance. We introduce a working deliberation prototype and a study design for assessing review quality, human time, and reviewers’ subsequent ability to evaluate new manuscripts without AI assistance. We invite venues and reviewer cohorts to collaborate in testing how AI-assisted workflows can meet immediate review demands while developing expertise for future cycles.