37 researchers propose AI-native format to replace traditional scientific papers

Key takeaway

• Thirty-seven researchers from leading universities and tech companies have published a provocative paper proposing that scientists abandon traditional papers in favor of an "Agent-Native Research Artifact" (ARA) format optimized for AI agents to read, reproduce, and extend research autonomously.

• Lead author Jiachen Liu, who earned her Ph.D. from the University of Michigan in 2025 and cofounded the Agent Native Research Lab in May, argues that AI agents are becoming autonomous contributors to research rather than tools, and that current papers waste roughly 80 percent of research knowledge by omitting failures and implementation details.

• The proposal addresses a centuries-old friction in scientific collaboration while acknowledging concerns about AI hallucination—Liu is working on formal systems combining neural and symbolic AI to ensure all claims are logically rigorous and verifiable.

3 Key Points

Context & Analysis

FAQ

What is an Agent-Native Research Artifact (ARA), and how does it differ from a traditional paper?

An ARA is a format designed for AI agents to use efficiently, presenting research in a way machines can directly read, reproduce, and extend. Unlike traditional papers, which the authors say lose roughly 80 percent of research information (including failures, decision-making, and fine-tuning details), an ARA includes a "Live Research Manager" that automatically observes and documents the entire research process without requiring manual effort from the scientist.

Who are the researchers behind this proposal, and when was it published?

Thirty-seven researchers from roughly two dozen top universities and tech companies published the paper in May. Lead author Jiachen Liu completed her Ph.D. in computer science from the University of Michigan in 2025 and cofounded the Agent Native Research Lab in Palo Alto, California, in May.

How will the system prevent AI from making errors or hallucinating?

Liu is working on a formal system using neurosymbolic techniques (combining neural networks with symbolic logic) that guarantees rigor by ensuring every claim is written as a mathematical formula and can be proved by the system. Rather than using one language model to supervise another, this formal approach makes all claims in the system self-consistent and objectively verifiable.