Stanford's Paper2Agent converts 74 of 100 papers to agents

Stanford's Paper2Agent converted 74 of 100 computational biology papers from bioRxiv into working AI agents, and 593 of the 599 tools those agents generated passed automated validation, Nature reports. The system reads a paper's text, code, datasets and supplementary files, deposits them on an MCP server, then lets AI agents autonomously build callable tools on top of that bundle.

Jiacheng Miao, Joe R. Davis, Yaohui Zhang, Jonathan K. Pritchard and James Zou ran the tool on the AlphaGenome paper, an AI model that predicts properties of DNA sequences. It produced an agent in about 45 minutes at a computing cost of US$14, and the agent answered genetics questions with near-perfect accuracy, outperforming Biomni, a competing biomedical AI agent.

"Knowledge should not be static records. It really should be dynamic and interactive," Zou told IEEE Spectrum, which reported that 22 tools were generated from the AlphaGenome paper alone and that a testing agent runs up to six diagnostic attempts per function. Zou also framed the conversion process itself as a documentation signal: "Agentification itself is a useful certificate that says, 'This work is relatively complete and well documented.'"

Dongping Chen, a computer scientist at the University of Maryland, called the idea "quite compelling" in the same piece. Three of the researchers we track shared the Nature story on the day it ran.