Artificial Intelligence & Scientific Research: Scientists at Stanford have developed a new artificial intelligence framework that can transform conventional research papers into interactive AI agents capable of explaining scientific work, applying published methods to new data and communicating with other research agents.

The system, called Paper2Agent, was described in research published in Nature on September 16, 2026. Instead of treating a scientific paper as a static document, the technology turns its research methods, code, datasets and supporting material into an interactive system.

From Static Papers to Interactive Knowledge

Scientific publications traditionally require researchers to read a paper, understand its methodology, locate associated software and data, and then reproduce or adapt the work themselves.

Paper2Agent is designed to reduce some of these barriers. The system analyses a manuscript together with its available code and data, then creates an AI-powered interface through which researchers can interact with the underlying research.

The resulting agent can function like a virtual corresponding author, answering questions about the study and helping users work with the methods described in the publication.

AI Agents Can Apply Published Methods

One of the notable capabilities demonstrated by the researchers is the ability to use a paper agent on new datasets.

Rather than simply summarising the conclusions of a study, the agent can expose tools based on the paper’s methodology. Researchers can then interact with those tools using natural-language instructions.

This could make it easier for scientists from different disciplines to understand and experiment with methods developed by another research group.

Papers Can Communicate With Each Other

The research also demonstrates a more advanced possibility: multiple paper-specific AI agents can communicate and work together.

This creates a new way of connecting scientific knowledge. Instead of researchers having to manually identify relationships between studies, AI agents representing different papers could potentially identify complementary methods and information.

The Stanford team demonstrated this concept by allowing paper agents to collaborate in research tasks.

Testing the System

The researchers tested Paper2Agent on several scientific publications and their associated computational resources.

One demonstration involved the AlphaGenome research, where the system converted the publication and its computational resources into an interactive agent. Nature reported that the AlphaGenome agent was created in roughly 45 minutes, with computing costs of about US$14 for that example.

The research team has also created more than 100 paper agents as part of its ongoing work on the technology.

Potential Impact on Scientific Collaboration

If systems like Paper2Agent become widely usable, scientific publications could become more than records of completed research. They could also become reusable computational resources.

Researchers might be able to ask questions about unfamiliar studies, test published methods on their own datasets and combine techniques from different fields without having to reproduce every technical step manually.

The technology could therefore support faster knowledge transfer and make computational research easier to reuse.

Human Researchers Remain Important

The researchers have also stressed the importance of attribution and oversight. AI agents extending the work of scientific papers should continue to identify and credit the original researchers and publications.

The team is also studying how large numbers of paper agents could interact safely and productively as the technology develops.

Paper2Agent represents an emerging approach in which AI does more than summarise scientific literature. By connecting published knowledge with executable methods and data, the system points toward a model of research in which scientific papers can become interactive tools for further investigation.