Incyte (Nasdaq: INCY) has entered a collaboration agreement with Edison Scientific, a San Francisco-based AI company founded in 2025, to embed Edison's Kosmos platform across its drug discovery and development workflows. The arrangement marks one of the first deployments of an agentic AI scientist system — one designed to execute multi-step research tasks autonomously rather than assist human researchers at individual decision points — within a mid-to-large biopharma R&D organization. Financial terms were not disclosed.
Kosmos will initially be applied to target discovery, target validation, and translational biology at Incyte, with the stated intention of expanding across the company's broader R&D organization over time. The platform is described by Edison Scientific as capable of processing approximately 1,500 full-length scientific papers and executing roughly 42,000 lines of code within a single research run, delivering structured scientific reports with source-traceable outputs.
Deal context
Edison Scientific was spun out of FutureHouse, a nonprofit AI research laboratory, and has raised USD 70 million from Spark Capital, Triatomic Capital, and other investors. Kosmos inherits architectural elements from FutureHouse's earlier systems, including PaperQA2, a retrieval-augmented generation framework for scientific literature, and Robin, a multi-agent system that autonomously identified ripasudil — a drug approved in Japan for glaucoma — as a candidate for dry age-related macular degeneration. That finding remains preclinical and has not advanced to clinical trials.
The platform operates through parallel agent instances: a data analysis agent that iteratively writes and executes code within a Jupyter notebook environment, and a literature search agent that reads full-length papers at scale. Both agents share information through what Edison Scientific describes as a structured world model — a continuously updated internal representation of the problem space that maintains coherence across heterogeneous data sources. The company reports that approximately 80% of statements in Kosmos-generated reports are verified as accurate by human scientists, a figure that implies meaningful but imperfect output quality requiring researcher review.