Stanford's Paper2Agent turns any research paper into an AI agent

Stanford researchers built a working AI agent out of the AlphaGenome paper in about 45 minutes, at a computing cost of US$14. It then answered genetics questions with near-perfect accuracy and outscored other biomedical AI agents given the same tests, including Biomni, a tool developed by academic researchers that draws from dozens of databases.

The system doing this is called Paper2Agent, described in Nature on 16 September 2026. It takes a paper's text, code, and datasets and deposits them on an MCP server. A team of AI agents then autonomously builds tools that apply the paper's methods to fresh data, and users query the result through the large language model of their choice.

James Zou, a computer scientist at Stanford University and co-author, framed it as a shift in what a published paper is for. The technology, he said, "can help us to reimagine what knowledge looks like in the future."

Each agent is pitched as a "virtual corresponding author": ask it what the paper did, apply its methods to your own data, or let it collaborate with agents built from other papers on cross-disciplinary questions. According to MarkTechPost's coverage, a batch run over 100 bioRxiv papers converted 74 of them, and three agents working together identified GPR137 as a probable causal gene for psoriasis. Three of the AI researchers we track had already circulated the link by the time it reached our feed.

Neither write-up says whether authors of the source papers are asked to consent before their work is agentified, or how the method fares outside biomedicine.