Systematic Comparison of Agentic AI Frameworks for Scholarly Literature Processing

Frameworks for agentic artificial intelligence (AI) are becoming popular as instruments for automating intricate processes, such as those related to academic research. Six popu- lar frameworks—AutoGen, Google ADK, CrewAI, LlamaIndex, LangGraph, and Semantic Kernel— were compared in this study, with an emphasis on their architectural features and suitabil- ity for literature processing tasks. We developed a prototype system using AutoGen to summarize preprints from arXiv to demonstrate its practical use. We analyze the interoperability of this system with other frameworks and describe how workflows are orchestrated within it. Although there are still issues with synthesis quality, citation accuracy, and scalability, our initial assessment suggests that agentic AI systems may enable wider source coverage and less manual labor in early stage literature review. The study contributes a taxonomy of framework design patterns, an initial demonstration of agentic workflows for academic tasks, and a discussion of open challenges for future research.