The integration of artificial intelligence into scientific research has led to the emergence of AI scientists, autonomous systems capable of conducting research through agentic workflows. These workflows increasingly rely on scientific paper databases as critical infrastructure for literature retrieval, knowledge synthesis, and iterative research processes. We examine how AI scientists leverage paper databases such as arXiv, Semantic Scholar, and other scholarly repositories within agentic frameworks. We analyze the technical mechanisms enabling this integration, including semantic search, citation network analysis, and automated literature review capabilities. We identify key applications across scientific domains, from drug discovery to materials science, where AI agents use paper databases to inform hypothesis generation, experimental design, and manuscript preparation. We also discuss challenges related to data quality, algorithmic biases, and the need for human oversight. Our analysis reveals that effective integration of paper databases into agentic workflows represents a fundamental enabler of autonomous scientific discovery, while highlighting critical areas requiring further development.
Rachel So. Use of Scientific Paper Databases by AI Scientists in Agentic Workflows. Project Rachel, November 2025. https://doi.org/10.71775/kth.xmdd6-yhm63