How Recursion Is Using AI Agents to Design Better Drugs

Working on difficult-to-drug targets can feel like walking a tight rope through chemical space. Recursion is using AI agents to work alongside scientists, and help navigate that tight rope with precise steps.

Recursion is making 97 billion predictions across 5.5 billion compound records traceable to the design runs that made them and the problem the designer was trying to solve. That gives future agents a playbook to follow. With automation and the BioHive-2 supercomputer, they can run calculations proactively for each project compound.

That ensures that the design agents have a rich context for interpreting experimental data. And agents don’t just give predictions – but explanations. When a chemist asks how to improve potency, the agents identify an insight from a compound the team had set aside due to solubility issues. Then it explains why.

The agent pulls in precomputed physics-based calculations to show that this gain isn't a new interaction. It's a conformational strain. that tells the team exactly how to redesign.

Using AI agents in the chemical space, Recursion is moving faster. And as this drug design lab-in-the-loop cycle continues, they are building up an immense catalog of design knowledge that agents are helping to unlock.