Edison Scientific

Edison Scientific Inc. is a San Francisco-based company that builds autonomous "AI scientist" agents for research and development. It is the commercial spinout of FutureHouse, a nonprofit AI biology lab founded in 2023, and it commercializes the agent research developed there. Its platform, available at platform.edisonscientific.com, combines literature synthesis, data analysis, molecular design, hypothesis generation, novelty assessment and experimental planning in a single environment intended to compress research tasks that normally take weeks or months.

The flagship product is Kosmos, an AI scientist that takes a research objective and one or more datasets, then autonomously reads literature, writes and executes analysis code, generates hypotheses and produces a comprehensive cited report. Kosmos orchestrates specialized agents in parallel and maintains a structured world model so it can reason across large datasets and literature collections; the company states a single run involves reading about 1,500 papers and executing about 42,000 lines of code, with every conclusion traceable back to the code or literature passage that produced it. Surrounding agents include Literature (search and synthesis over a corpus of more than 175 million papers, trials and patents, with handling of citation graphs, journal quality and clinical trial data), Analysis (flow cytometry, RNA-seq and other experimental data, producing statistics and publication-ready figures), Precedent (novelty and prior-art assessment across fields), and Molecules (chemistry-focused molecular design and analysis).

The company markets the platform to biopharma R&D teams for workflows spanning target identification through clinical development and filing, and positions it as model-agnostic across frontier models. Its own site lists a research lineage of ChemCrow (2022, language agents for chemistry), PaperQA (2023, deep research agent), Aviary (2024, trained AI agents for science) and Robin (2025, described as validated scientific discoveries via an AI agent in biology published in Nature).

Founding story

FutureHouse, a nonprofit AI biology lab, began in 2023 with the goal of building an AI scientist capable of accelerating innovation. Edison Scientific was created as its commercial spinout to bring that technology to market at scale, while FutureHouse continues work on basic biology and educational access. Co-founders are CEO Sam Rodriques, a physicist and bioengineer who worked on spatial and temporal transcriptomics, brain mapping, gene therapy and nanofabrication and briefly ran a lab at the Francis Crick Institute, and CTO Andrew White, a chemistry and machine learning researcher who led ChemCrow, ether0 and PaperQA.

Business model

Edison sells access to its AI agent platform, with a free tier for academics and paid subscriptions for users needing higher rate limits or additional features. For enterprises it offers on-premise deployment inside the customer's environment, custom "world models" built on customer data, and forward-deployed scientists embedded with customer teams; commercial engagement typically starts with a booked demo. It also enters partnerships with pharmaceutical and biotech organizations, such as Incyte Corporation and Population Health Partners.

Subscription/paid plans layered on a free academic tier, plus enterprise deployments that bundle software with embedded scientific staff and custom model development. Specific pricing is not disclosed in the sources.

Traction

Roughly 30 employees as of December 2025, with hiring planned across AI, biology, chemistry and engineering. A $70 million seed round was raised in December 2025 at a reported ~$250 million valuation. Reported early projects include identifying therapeutic targets, mapping complex molecular structures and narrowing candidate molecules. Announced 2026 partnerships with Incyte Corporation and Population Health Partners. Customer testimonials include Incyte EVP/Chief Scientific Officer Patrick Mayes and MIT-affiliated professor Tonio Buonassisi, who stated Kosmos executed roughly 292 independent analysis trajectories within 24 hours and surfaced insights his team had missed. Beta users estimated one Kosmos run equates to about six months of PhD or postdoc work.

Latest developments

Following its December 2025 seed round, the company planned to hire across AI, biology, chemistry and engineering, make Kosmos more autonomous, deepen integrations with lab information systems and collaboration tools, and expand assistant-style agents for tasks such as data analysis and precedent search. Its site subsequently announced the LABBench2 benchmark (February 2026), a partnership with Incyte Corporation (May 2026) and a partnership with Population Health Partners (June 2026).