Chai Discovery's founders on designing drugs in 24 hours and the Eli Lilly partnership
Key takeaways from a Sachin & Adam Show interview with Chai Discovery co-founders Josh MeierDealroom has a profile for this one. Try Dealroom → and Jack DentDealroom has a profile for this one. Try Dealroom → (ex-OpenAI) (July 2026):
From OpenAI to molecules. Jack Dent worked on early language models (GPT-1/GPT-2) at OpenAI and reasoned that if models worked for human language, they should work for the 'natural language' of DNA and proteins — the origin of Chai's approach.
Compressing drug discovery to 24 hours. The founders describe collapsing a process that took months or years into ~24 hours of compute plus a couple of weeks of lab testing — aiming ultimately at a 'zero-shot' drug candidate that comes out of the model ready for patients with no changes.
Chai-3 leap. Antibody design success rates went from ~0.1% in the lab a year ago, to 15-16% with Chai-2, to more than double that with Chai-3 — and, crucially, the molecules now come out with many drug-like properties already dialled in, not just as starting points.
Atomic-level accuracy. Structures predicted on-screen matched cryo-electron-microscopy lab results with an error smaller than the diameter of an atom; antibody-antigen prediction accuracy has improved ~10x over prior state-of-the-art AlphaFold.
Eli Lilly partnership. Chai is working with Eli Lilly, which stress-tests the models at scale across many projects and feeds results back — helping Chai set its research roadmap and stay customer-oriented.
Lean, in-person team. Roughly 30 people in San Francisco, interdisciplinary, few meetings, working at 'the pace of science'; progress comes from compounding 5-10% weekly improvements rather than one secret.
From open-source R&D to commercial. Chai open-sourced early work (not yet valuable enough to charge for); with Chai-2 it crossed the inflection point into a commercial, product-led org, with a flywheel where better models unlock more projects and revenue.
The real competitor: yeast. Meier frames the incumbent Chai must beat not as other AI labs but existing biological methods — with each release, the true competitor becomes the previous Chai model.
