Medra Says Its AI Can Run Drug Experiments. One Proof Point Exists. Nobody Can See It.
Genentech would not confirm the result Medra points to as proof its AI scientist works. The company wont say why — or share the data that would let anyone check.
When type0 asked Genentech to confirm the result Medra points to as proof its AI scientist works — antibody binding reported to have jumped from zero to over 70 percent after a protocol change — the company declined.
Genentech disclosed a partnership with Medra in December. It would not confirm the antibody result.
The refusal is notable because the result is the closest thing Medra has to a proof-of-concept for its core technology: a system of roughly 100 robotic arms in a 38,000-square-foot San Francisco laboratory, equipped with cameras and nine sensors, that runs experiments continuously and logs every detail of how each one was performed. The company raised $52 million in a Series A last month, grew from 15 to 45 employees in five months, and has five customers with experiments scheduled. Founder Michelle Lee has described Medra as a TSMC for drug discovery — a manufacturing layer that lets biotech companies run experiments without building labs of their own.
What Medra cannot produce is the raw data that would let anyone check whether its AI scientist actually delivered the result Lee describes. The company will not identify the customer who ran the experiment, name the antibody target, or provide the measurements taken before and after the suggested change. A Medra spokesperson said proprietary constraints prevent sharing experimental data with outside parties. The claim appears only in a podcast interview with Lee. It is not in Medra's Series A announcement, its company website, or its profile in Genetic Engineering News.
Independent verification would require the customer name, the antibody target, and measurements before and after the step, confirmed by someone without a financial interest in the outcome. Medra has not provided any of those things.
What Medra has built, instead, is a machine for accumulating process knowledge. Every pipette angle that produces a good result, every vortex duration that changes a yield, every timing variation between reagent additions — the system logs it automatically. Lee describes this as the compounding edge: the more protocols the company runs, the denser and more useful the underlying dataset becomes. Customers own their experimental data — sequences, targets, candidates. Medra retains the procedural knowledge of how each experiment was run. More than 85 percent of Medra's customers arrive with requests the company has never fulfilled before; the process knowledge gained from each one compounds.
"Science is so critical to the United States' — any nation's — prosperity and also national security," Lee told the podcast. "If all our antibiotics come from abroad, what happens when there's a national security crisis?" It is an argument for domestic scientific infrastructure at a moment when Chinese pharmaceutical manufacturing has accumulated process knowledge at a volume no American lab has matched.
The arms are still running in San Francisco. Whether they proved what Lee says they proved is a question she has not yet answered in a way that anyone outside her company can check.
