Stanford Paper2Agent turns 74 of 100 bio papers into agents

Stanford researchers have built a system that converts a research paper into a callable AI agent, and applied it at enough scale to say how often it works. Paper2Agent, described in Nature on 16 September 2026 by Jiacheng Miao, Joe R. Davis, Yaohui Zhang, Jonathan K. Pritchard and James Zou, wraps a paper's manuscript, code, data and workflows behind a Model Context Protocol server that a chat agent such as Claude Code can call. The pitch is that the paper becomes a virtual corresponding author: ask it what the work did, apply its methods to new data, or hand off to agents built from other papers.

The AlphaGenome case study was generated in roughly 45 minutes at about $14 of compute on a personal laptop, produced 22 tools, and scored 98.7 ± 1.3% on 15 tutorial-derived queries. Run at scale on 100 computational biology papers from bioRxiv, the pipeline agentified 74, producing 599 tools of which 593 passed automated validation. Across 300 benchmark questions, the resulting agents averaged 91.2 ± 1.6% accuracy at roughly $0.20 a query.

The paper frames the shift bluntly. Paper2Agent "transforms research output from passive artefacts into active systems that accelerate use and discovery," the authors write, exposing manuscript, supplementary materials, datasets, code and workflows "as active, agent-native knowledge rather than static text."

James Zou, the Stanford computer scientist who co-authored the work, told IEEE Spectrum that the by-product is a kind of quality signal: "Agentification itself is a useful certificate that says, 'This work is relatively complete and well documented.'" Of the 100 papers tried, 26 failed the pipeline, which the authors attribute to incomplete code, missing documentation or incompatible software packages. In a demonstration that agents built from separate papers can collaborate, the team reports identifying GPR137 as a probable causal gene for psoriasis.

Five researchers we track in our Who's Who directory circulated the Nature link in the days after publication.