Summary
In this Latent Space podcast episode, hosts Brandon (of Atomic AI) and R.J. Honicky (CTO of Mirror Omix) sit down with Matt McPartlon and Neil Patil — co-founder and platform lead, respectively, of Chai Discovery — to discuss how AI is transforming drug discovery from a slow waterfall process into something closer to agile software development. Chai, now roughly 2.5 years old, has built a molecular design suite that resembles Figma or Photoshop for biology: users can load molecules, “paint” epitopes, and generate binders via AI in a visual interface, replacing months of wet-lab iteration with computational candidates.
The conversation covers Chai’s trajectory from Chai 1 through Chai 2 — the model that crossed the threshold of commercial usefulness and led to partnerships with Eli Lilly, Pfizer, Novartis, and GenX — and the mad buildout that followed to productize the models and secure sufficient compute to serve pharma partners. A recurring theme is the IP-security challenge: convincing notoriously data-sensitive pharmaceutical companies to run proprietary target sequences through a shared platform required building strict data isolation from the ground up.
Technically, the founders distinguish structure prediction (where held-out ground truth enables clean benchmarking) from molecular design (where validation requires physical lab synthesis), and explain how Chai’s design suite is climbing levels of abstraction — from basic binding prediction toward agonists, bispecifics, and antibody-drug conjugates — as models improve. Essential listening for anyone tracking AI’s role in biotech and the commercialization of foundation models in science.
📺 Source: Latent Space · Published August 11, 2026
🏷️ Format: Podcast
