Today, we announce the first discovery made by Robin, our multi-agent system for automating scientific research.
Read the preprint here: https://arxiv.org/abs/2505.13400
Robin code, data and full agent trajectories will be released Tuesday 5/27
FutureHouse's mission is to automate scientific discovery. Until now, we've released specialized AI agents to automate individual aspects of the discovery process: Crow, Falcon, and Owl for literature search and synthesis; Phoenix for chemical synthesis design; and Finch for complex data analysis.
Today, we are excited to announce a significant breakthrough: by integrating these agents into a unified system named Robin, we've been able to automate the key intellectual steps of the entire scientific process and achieve our first AI-generated discovery. Robin is a workflow that orchestrates Crow, Falcon, and Finch to propose and pre-clinically validate novel treatments for any human disease. We applied Robin to identify ripasudil, a Rho-kinase (ROCK) inhibitor clinically used to treat glaucoma, as a novel therapeutic candidate for dry age-related macular degeneration (dAMD), a leading cause of irreversible blindness worldwide.
How Robin Made Its First Discovery
Robin made this discovery through an iterative cycle of hypothesis generation, experimental design, and data analysis:
- Initial Hypothesis: Robin used Crow to conduct a broad literature review, and hypothesized that enhancing retinal pigment epithelium (RPE) phagocytosis could provide therapeutic benefit for dAMD. Robin then used Falcon to evaluate a set of candidate molecules that might be able to do this, and we tested ten of them in the lab. Using Finch, Robin then analyzed the data from these experiments to find that the ROCK inhibitor Y-27632 augmented RPE phagocytosis in cell culture.
- Mechanism Investigation: Robin next proposed an RNA-sequencing experiment to determine if Y-27632 was inducing gene expression changes that might explain the increased phagocytosis of RPE cells. We did the experiment, and Finch analyzed the data and identified that Y-27632 upregulated ABCA1, a critical lipid efflux pump in RPE cells.
- Discovery of ripasudil as a novel dAMD treatment: Using the data from the first round of drug candidate testing, Robin proposed a second set of drug candidates. We tested them in the same experiment to find there was a new top hit: ripasudil, a drug already used in the eye.
All hypotheses, experiment choices, data analyses, and main text figures in the manuscript describing this work were generated by Robin autonomously. Human researchers executed the physical experiments, but the intellectual framework was entirely AI-driven.
A New Paradigm for Scientific Research
What's particularly notable is how quickly we were able to build Robin and use it to make this discovery. The entire process—from conceptualizing Robin to paper submission—was completed in just 2.5 months by a small team of researchers. Our goal was to determine whether it was possible to generate and validate a novel therapeutic idea using an interpretable workflow of only three agents. We think we’ve achieved that.
By automating hypothesis generation, experimental planning, and data analysis in an integrated system, Robin represents a powerful new paradigm for AI-driven scientific discovery. Although we first applied Robin to therapeutics, our agents are general-purpose and can be used for a wide variety of discoveries across diverse fields—from materials science to climate technology. We're releasing Robin as open-source on Tuesday, May 27th, and hope that our approach of orchestrating our agents in simple workflows will inspire others to build their own systems for automated discovery.
