FutureHouse AI Agents Accelerate Scientific Discovery by Automating Research Workflows

FutureHouse Develops AI Agents to Automate Scientific Progress

FutureHouse, co-founded by Sam Rodriques PhD ’19, has created AI agents that automate crucial steps in scientific research. Their goal is to address the slowdown in scientific productivity that has been observed over the past 50 years, where discoveries now require more time, funding, and larger teams.

As research grows more specialized and complex, scientists spend increasing amounts of time reviewing literature, designing experiments, and analyzing data. FutureHouse’s AI platform tackles these bottlenecks by automating tasks like information retrieval, synthesis, chemical design, and data analysis.

Why AI Agents for Science?

Rodriques points out that natural language is the primary way scientific discoveries are communicated, hypothesized, and reasoned about. While some AI models focus on biological data like DNA or proteins, FutureHouse focuses on building agents that understand and generate scientific knowledge through natural language.

Rodriques’ experience during his PhD at MIT highlighted a critical problem: even if all the data about brain function were available, no one has the time to read and integrate all the relevant literature to form comprehensive theories. This insight laid the foundation for FutureHouse.

From Concept to Platform

The idea to automate and scale scientific research grew from Rodriques’ interest in large research collaborations and new organizational structures to boost productivity. After encountering large language models like Chat-GPT 3.5 and 4, Rodriques teamed up with Andrew White, a computational chemist who had developed one of the first large language agents for science.

Initially, FutureHouse built separate AI tools for literature searches, data analysis, and hypothesis generation. Their first public release, PaperQA (now called Crow), launched in September 2024, is regarded as one of the most effective AI agents for retrieving and summarizing scientific papers. Another early tool, Has Anyone (now Owl), helps researchers check if specific experiments or hypotheses have been explored.

Integrated AI Agents for Scientific Workflows

FutureHouse’s platform, officially launched on May 1, 2025, includes several agents designed to work together:

• Crow: Summarizes and retrieves scientific literature.
• Owl: Checks if experiments or hypotheses have been previously explored.
• Falcon: Reviews and compiles more sources than Crow.
• Phoenix: Assists in planning chemistry experiments with specialized tools.
• Finch: Automates data-driven discovery in biology.

On May 20, FutureHouse demonstrated a workflow that automated key scientific steps to identify a potential therapeutic candidate for dry age-related macular degeneration (dAMD), a disease causing irreversible blindness. The company also released ether0, a 24-billion-parameter open-weights reasoning model for chemistry, in June.

Rodriques emphasizes that these agents function best as parts of a connected system, working seamlessly from literature search through data analysis and experiment planning.

Accessible AI Agents for Researchers

FutureHouse’s AI agents are publicly available at platform.futurehouse.org. Early users report accelerated research outcomes. For example, one scientist identified a gene linked to polycystic ovary syndrome and proposed a new treatment hypothesis. Another researcher at Lawrence Berkeley National Laboratory used Crow to build an AI assistant that searches PubMed for Alzheimer’s disease research.

Additionally, FutureHouse’s agents have outperformed general AI tools in systematic gene reviews related to Parkinson’s disease.

Rodriques suggests users who treat these agents as intelligent research assistants will gain the most. For faithful literature reviews, FutureHouse’s agents outperform general-purpose AI models, which might be better suited for speculative queries.

Looking Ahead in Scientific AI Automation

FutureHouse aims to enhance its agents so they can analyze raw research data to test reproducibility and validate conclusions. Longer term, they plan to embed tacit scientific knowledge into the agents. This will allow more sophisticated analyses and enable the AI to use computational tools to explore hypotheses.

To keep advancing science, FutureHouse is working on integrating their agents with foundation models for biology and providing access to specialized scientific tools. This infrastructure will help agents perform a broader range of tasks essential for scientific discovery.

For researchers interested in AI tools that support scientific workflows, exploring FutureHouse’s platform offers a practical way to increase efficiency and tackle complex problems faster.