Survey of Autonomous Research Agents Framework

This document is a technical report that surveys the emerging field of autonomous research agents, which are AI systems capable of conducting research independently. It outlines a five-level autonomy taxonomy that categorizes systems from basic code assistance to fully autonomous research agents. The report analyzes four dominant architectural patterns used in these systems, including single-agent loops and multi-agent collaboration, and evaluates their trade-offs in terms of scalability, cost, reliability, and human oversight. Additionally, it provides a detailed analysis of 17 major systems, revealing that most operate at Level 4 autonomy, while Level 5 remains aspirational. The report identifies six fundamental open problems in the field, such as reproducibility and safety, and proposes research directions to address these challenges. The findings highlight the significant implications of autonomous research agents for academic practices, intellectual property, and the future of scientific inquiry.