FutureHouse is a non-profit AI research organization based in San Francisco that builds AI agents to automate scientific research, first in biology and ultimately across what it calls "other complex sciences." It describes itself as a philanthropically funded moonshot with a stated 10-year mission to build an "AI Scientist": semi-autonomous AI systems for scientific research intended to accelerate the pace of discovery and to widen worldwide access to scientific, medical, and engineering expertise (Source: futurehouse.org).
Overview
| Field | Value |
|---|---|
| Type | Non-profit research organization |
| HQ | San Francisco, USA |
| Founders | Sam Rodriques (CEO, co-founder), Andrew White (Head of Science, co-founder) |
| Funding model | Philanthropic; associated with Eric Schmidt's science-philanthropy efforts |
| Focus | Autonomous and semi-autonomous AI agents for scientific discovery (biology, chemistry, drug discovery) |
FutureHouse pursues end-to-end automation of the research loop rather than AI as an assistive tool within a human-run process, placing it within the AI for Science area. As a nonprofit rather than a for-profit frontier lab, it is filed in entities/. Its work figures in debates over scientific validity, reproducibility, authorship, and the regulatory treatment of AI-originated discoveries in drug development, and connects to the agentic-AI thread, since systems such as Robin involve autonomy over a multi-step workflow rather than a single-model capability claim.
Activities
The FutureHouse Platform is a suite of specialized scientific agents, including literature-search and data-analysis agents that the organization names Crow, Falcon, and Finch, offered to scientists via a web interface as "superintelligent scientific agents."
Robin, announced in May 2026, is a multi-agent system that the organization describes as the first to fully automate both hypothesis generation and data analysis for experimental biology. Robin orchestrates the platform's literature-search and data-analysis agents to propose hypotheses, design experiments, interpret results, and revise hypotheses in an iterative "lab-in-the-loop" cycle. Applied to dry age-related macular degeneration (dAMD), Robin proposed enhancing retinal-pigment-epithelium phagocytosis and identified ripasudil, a clinically used ROCK inhibitor not previously proposed for the disease, confirming in-vitro efficacy. The work was published in Nature on 2026-05-19 (Robin: a multi-agent system for automating scientific discovery (Nature, May 2026)). Robin is a claimant to having moved AI from research assistant to research agent.
The paper's provenance claim is the basis for that characterization: "All hypotheses, experimental directions, data analyses and data figures in the main text of this report were produced by Robin." The architecture uses three named agents — Crow and Falcon for concise and deep literature search, Finch for data analysis — with humans running the physical experiments, making the system "semi-autonomous" by design. Following the ripasudil result, Robin proposed and analysed an RNA sequencing experiment revealing upregulation of ABCA1, "which encodes a lipid efflux pump and represents a possible novel target" (Robin: a multi-agent system for automating scientific discovery (Nature, May 2026)).
The authors characterize the advantage as "combinatorial synthesis (identifying non-obvious connections between disparate fields)," targeting "low-hanging fruit that human experts may overlook due to the compartmentalization of scientific knowledge," motivated by historical repurposing lags — dabrafenib's otoprotective effects found a decade after its mechanism was characterized, ketamine at 22 years, KarXT at 13. They also state limits: Robin "does not yet produce precise, executable protocols," Finch "is reliant on prompt engineering by domain experts," the results used frontier models available in early 2025, and the dAMD hypothesis "would of course require validation in a suitable disease model and ultimately in a randomized, placebo-controlled trial." Four safeguards are described, including reliance on off-the-shelf models' existing safety alignment "to prevent the generation of malicious biological protocols" and an LLM classifier screening queries against unsafe topics (Robin: a multi-agent system for automating scientific discovery (Nature, May 2026)).
Other releases include DISCO (April 2026), described as AI design of enzymes for chemistry "that nature never explored"; OXtal (December 2025), a diffusion model for molecular crystal-structure prediction; and Edison Scientific, a related entity announced in November 2025 and spun out to commercialize the organization's scientific-AI work.
Relationships
- related: AI for Science, Agentic AI, Healthcare — AI Deployment, AI in Healthcare
- related: Eric Schmidt (science-philanthropy backing)
- deploys-in: drug discovery / experimental biology
Sources
- FutureHouse organizational site — futurehouse.org
- Ghareeb, Chang, Mitchener et al., "A multi-agent system for automating scientific discovery," Nature, 2026-05-19 — nature.com
Page created 2026-05-24 by the gap-identifier (recent-thread gap from the 2026-05-23 developments log; no prior wiki page existed for the organization).