AgenticScience/Awesome-Agent-Scientists

What it solves

Scientific discovery is moving from humans using AI as tools to AI agents that can autonomously formulate hypotheses, design experiments, and refine theories. However, prior work is fragmented, focusing on separate aspects like workflows, autonomy, or architecture, and often neglects the natural sciences. This project bridges that gap by providing a unified framework and the first domain-oriented review of agentic science.

How it works

The repository, part of the Intern Discovery project, organizes a comprehensive survey as a structured list of papers and resources. It introduces a framework that identifies five core capabilities of scientific agents: reasoning and planning, tool use, memory, multi-agent collaboration, and optimization/evolution. It also defines a four-stage dynamic workflow: observation and hypothesis generation, experimental planning and execution, data interpretation and analysis, and synthesis and iterative refinement. The content is partitioned into parts—a fully autonomous research pipeline and then systematic reviews across life sciences, chemistry, materials science, and physics, each with subcategories.

Who it's for

This is for researchers, AI/ML practitioners, and natural scientists who want to understand the landscape of AI agents for scientific discovery, identify relevant frameworks, or get inspired for their own work. It also serves as a reference for students and educators in interdisciplinary fields.

Highlights

• First domain-oriented review of autonomous scientific discovery, spanning life sciences, chemistry, materials, and physics.
• Anatomy of scientific agents into five core capabilities: reasoning/planning, tool use, memory, multi-agent collaboration, and optimization/evolution.
• Dynamic workflow model with four stages: observation/hypothesis, experiment planning/execution, data interpretation, synthesis/refinement.
• Curated list of papers and projects, from general-purpose pipelines to domain-specific agents.
• Includes a companion paper on arXiv and an official homepage, with citation support.