It buys the drugs big pharma left on the shelf - then runs them through a machine built of software.
Walk into Formation Bio's Manhattan office and you will not find beakers. You will find a whiteboard covered in trial arithmetic, a few people arguing about it, and a quiet conviction that the slowest part of medicine has nothing to do with chemistry. The drugs already exist. The problem is everything that happens after.
Formation Bio is an AI-native pharmaceutical company headquartered in New York. It does not run a discovery lab hunting for new molecules. Instead, it acquires and in-licenses clinical-stage drug candidates - compounds that have already shown a pulse - and pushes them through a development engine made largely of software. The pitch is unfashionably literal: take a promising drug, run its trial faster and cheaper than anyone else, and hand it off to a larger company that wants the finished asset.
Here is the uncomfortable arithmetic of modern medicine. Bringing a single drug to market routinely runs into the billions of dollars and takes the better part of a decade. Most of that time is not spent discovering anything. It is spent on logistics - writing protocols, recruiting patients, managing sites, filing regulatory paperwork, watching for side effects, and cleaning data that arrives in a hundred incompatible formats.
CEO Benjamine Liu learned this the slow way. As a doctoral student, he was using machine learning to chase treatments for neurodegenerative diseases like Parkinson's and Alzheimer's. The science was thrilling. The development was not. He noticed that good drug candidates kept stalling, not because they failed, but because the machinery around them moved like it was 1995. The insight that built the company was almost anticlimactic: it wasn't discovery that held progress back - it was development.
This is the central tension the whole company is built around. Biotech, as an industry, is obsessed with the molecule. The molecule is the romance. The trial is the chore nobody wants to talk about at conferences. Formation Bio's wager is that the chore is where the years - and the billions - actually hide.
The company began in 2016 as TrialSpark, founded by Benjamine Liu, Linhao Zhang, and Kit Dobyns. Zhang and Dobyns had met years earlier as roommates during a Salesforce internship - the kind of friendship that survives years of arguing about hard problems, which turned out to be useful when the problem was an entire industry's operating model.
Their early move was almost cheeky in its simplicity: in 2018 they put clinical trial sites directly inside physicians' offices, meeting patients where they already were instead of building expensive standalone centers. Dobyns left in 2019 to found a maternity-care company. Liu and Zhang stayed and kept widening the thesis - from running better trials to owning the drugs being trialed. In December 2023, TrialSpark became Formation Bio. The rename was less a rebrand than a confession: this was no longer a trials vendor. It was a pharma company.
Acquire or in-license a clinical-stage drug. Run its trial through an AI-enabled platform faster and cheaper than traditional pharma. Then out-license or sell the de-risked asset to a bigger company. Each program lives in its own subsidiary - a hub-and-spoke structure that keeps the bets separate and the upside clean.
Formation Bio's technology is not one shiny tool but a stack of unglamorous ones, each aimed at a different stage of the slog. The AI-Driven Drug Development layer automates medical writing, protocol design, regulatory submissions, and biostatistics. The Clinical Trial Engine handles recruitment, site management, and compliance. Underneath both sits a Universal Data Platform that drags clinical, regulatory, and external datasets into one place so the AI has something coherent to read.
Automates medical writing, protocols, regulatory submissions, biostatistics, and safety monitoring.
Patient recruitment, site management, data management, and compliance in one system.
One knowledge base consolidating clinical, regulatory, and external data for the models to use.
Turns a trial protocol into tailored patient recruitment materials in roughly 15 minutes.
The least boring example is Muse, built with OpenAI and Sanofi. Feed it a dense trial protocol and it produces tailored patient recruitment materials in about 15 minutes - a task that used to involve a small committee and a long week. Elsewhere, the company says AI cut the time to draft an adverse-event report from roughly ten hours to one. Unglamorous? Completely. Ten hours to one is also the kind of math that, multiplied across a trial, turns into months.
Liu, Zhang, and Dobyns set out to fix how trials are run.
Sites embedded where patients already are.
Valuation crosses $1 billion.
The rename signals the shift from trials vendor to pharma company.
A first-of-its-kind deal to build purpose-built AI agents for drug development.
Led by a16z and Sanofi; a "material step up" from the $1B valuation.
Subsidiary Libertas Bio licenses gusacitinib to Sanofi for up to €545M.
Figures as reported by the company. Treat as directional, not audited - the point is the order of magnitude.
Skeptics are right to ask whether any of this is more than a deck. The answer arrived in June 2024, when Formation Bio closed a $372 million Series D led by Andreessen Horowitz and Sanofi, with Sequoia, Thrive Capital, Emerson Collective, and others joining. The company declined to name a new valuation but called it a "material step up" from its $1 billion Series C mark. Scott Kupor of a16z took a board seat.
The Sanofi relationship is the tell. A month before the raise, Sanofi, OpenAI, and Formation Bio announced a collaboration to fine-tune models on Sanofi's proprietary data and build AI agents for drug development - described as a first of its kind in the industry. By 2025, Formation's subsidiary Libertas Bio had licensed the drug gusacitinib back to Sanofi for up to €545 million. When one of the world's largest drugmakers is simultaneously your investor, your AI partner, and your customer, the thesis has stopped being theoretical.
Formation Bio is most often compared to Roivant Sciences, which pioneered a similar buy-and-develop, hub-and-spoke model. Others in the orbit: Lindus Health (the self-styled "anti-CRO"), Tempus, Insilico Medicine, and the traditional contract research organizations whose lunch this whole category is trying to eat.
Strip away the funding rounds and the model is about one thing: patients waiting. Every month shaved off a trial is a month a treatment reaches someone sooner. Formation Bio frames its mission simply - to bring new treatments to patients faster and more efficiently. The CNBC Disruptor 50 nod in 2025 was recognition that the boring middle of pharma might be the most valuable thing to disrupt.
There is an irony the company seems comfortable with. It is trying to make drug development exciting by making the exciting parts irrelevant - automating the writing, the recruiting, the filing, until the only thing left to be brilliant about is which drug to back. That is a strange ambition for a pharma company. It is a very natural one for a software company that happens to own drugs.
Return to that Manhattan whiteboard. Three people, a column of numbers, a drug whose fate depends on how fast and how cheaply the next two years can be run. For most of pharma's history, that math was done by hand, slowly, at enormous cost, and a good number of viable drugs quietly died in the gap between "promising" and "proven."
Formation Bio's bet is that the gap is closing - that the column of numbers gets solved by an engine, that the two years become eighteen months, that a drug which would have stalled instead reaches a patient. The beakers were never the point. The whiteboard was. And the company's entire reason for existing is to make the slow part of that whiteboard disappear.