Generative is different from predictive
The phrase “generative AI drug discovery” is often used interchangeably with AI drug discovery in general, but it means something specific. Predictive AI answers a question: given this molecule, how likely is it to bind to this target? Generative AI creates something new: it designs molecules that have never existed, optimized for a target from the start. The difference is the difference between ranking a list and writing a new one.
That distinction matters for the “sub-$10 AI drug stock” promo from Alexander Green’s Oxford Club. The company behind the tease, Absci, is a generative AI shop in the literal sense: it converts proteins into a digital form and uses generative models to design new antibody candidates. Understanding what “generative” actually means here is the key to understanding why the pitch is both real and overstated.
How a generative model designs a protein
A generative model is trained on a large number of examples, in this case proteins and their structures, and it learns the patterns that make a protein fold and function. Once trained, it can sample from that learned space and produce new sequences that match the patterns but are not copies of anything in the training data. For drug discovery, the model is steered toward a target, and it proposes designs most likely to bind to it.
That is the mechanism behind Absci’s platform, and it is a genuine advance. Instead of screening existing libraries, the model generates candidates on demand and ranks them, which collapses the search from physical testing to computation. We cover the full mechanism in our AI powered drug discovery explainer.
The gap between design and drug
The honest caveat is that a generated design is a candidate, not a drug. A model can propose a protein that looks perfect on paper, and the molecule can still fail in a human body for reasons the model never saw: it might not be stable, it might trigger an immune response, or it might not reach the target in a living organism. The generative step got faster, but the steps after it, the laboratory validation and the clinical trials, still run on biological time.
That gap is where promotional language tends to overreach. A pitch that describes a generative platform as the future of medicine is describing a real technology, but the distance between a generated design and an approved drug is still measured in years and dollars. Our generative AI in pharma piece covers how the big drugmakers are actually using these tools.
What generative AI is genuinely good at
Where generative AI earns its keep is in exploring space that brute-force screening cannot. The number of possible protein sequences is so large that no physical process could ever test a meaningful fraction of it. A generative model can search that space computationally and surface designs a human team would never have considered. That is a real advantage, and it is the honest core of the thesis.
Absci’s lead asset, ABS-201, is the concrete output of that process: an antibody engineered for an extended half-life, targeting pattern hair loss, with proof-of-concept data expected in the second half of 2026. It is the difference between a platform that can generate candidates and one that has moved one into trials. For the full picture of how the technology changes drug discovery, see our AI drug discovery explainer.
From image models to protein models
The technique behind generative drug discovery is not new; it is an adaptation of the same class of models that generate images and text. A diffusion model, the kind that produces a picture from a text prompt, can be retrained on protein structure data and asked to produce a new protein instead of a new image. The math is similar; the training data is different.
That lineage is why the field moved so fast. The tooling was already mature when the biology data caught up, and companies like Absci could borrow a decade of machine learning progress and point it at a harder target. It also explains why the bottleneck is not the model but the data and the wet-lab work that validates what the model proposes.
The bottom line
Generative AI drug discovery is a real and specific capability, not a marketing term. It turns drug design from a search through what exists into a search through what could exist, and that is a meaningful change. But a generated molecule is a starting point, not a finish line, and the companies that matter are the ones that have carried a generated design into human testing. Absci is one of them, and its antibody data due in 2026 is the test that will tell investors whether the generative thesis is more than a promise.
Ready to see the research? Click here to access Alexander Green’s report.
NewsletterVetter is an independent publication. We receive compensation from some of the services we review through affiliate links. Nothing on this site is investment advice. Always do your own research.