A Wall Street research shop just tried its hand at drug discovery. It brought an AI model instead of a lab coat.
Citrini Research, working with Anthropic, used the Claude Science model to hunt for potential scorpion antivenom candidates. The experiment produced roughly a year’s worth of early-stage design work in a matter of days. Then it produced something analysts care about even more: an investment thesis.
From a 35-page prompt to lab results
Citrini’s analysts fed Claude Science a detailed prompt of ~35 pages. Out came approximately 180 candidate molecules aimed at neutralizing scorpion toxins.
According to the research findings, work of that scope would normally take a year using traditional methods. Citrini compressed it into a short window.
That is where Twist Bioscience came in. The lab synthesized and tested 211 designs. Of those, 17 successfully bound to the target toxins.
Seventeen hits might not sound like a blowout. In early-stage drug discovery, where most candidates fail, it was enough to beat Citrini’s own expectations.
The bottleneck moves from software to the lab
On October 20, 2026, Citrini published a report laying out what it learned. The core argument is less about the AI itself and more about what happens after the AI finishes its homework.
If computational design gets dramatically faster, the slow part of the process shifts. The constraint is no longer coming up with candidates. It becomes testing them.
Citrini’s report frames this as a shift in the drug development story. Attention moves away from computational design tools and toward physical experimental validation and the lab infrastructure that supports it.
Citrini’s thesis is that these firms are likely to see rising demand for experimental validation services. Its report singled out two names as top picks: Charles River Laboratories International Inc. and Revvity Inc.
Part of a bigger pharma push into AI
Novo Nordisk struck a partnership with Anthropic in September 2026. That deal signaled that large pharmaceutical companies are taking AI-assisted research seriously.
The Citrini test adds a different kind of data point. It came from outside the pharma establishment, from analysts whose day job is picking stocks. If a research firm can generate validated binders with a long prompt and an AI model, the barrier to early-stage design looks lower than it used to.
What this means for investors and the industry
The obvious AI trade is to buy the companies building the models. Citrini’s report argues for a less obvious one: buy the businesses that sit downstream of the models and absorb the extra work they create.
There are reasons to keep expectations measured. The Citrini result covers a single target, scorpion antivenom toxins, and binding to a toxin is an early milestone. It is a long way from a binder in a lab to a treatment in a hospital.
The findings also note that a heavier emphasis on physical validation adds complexity to the supply chain. More molecules mean more synthesis, more testing and more coordination between design teams and labs.
Disclosure: This article was edited by Diego Almada Lopez. For more information on how we create and review content, see our Editorial Policy.