Molecule manufacturability and lab automation constrain AI drug development

Enveda’s CEO says the company uses NVIDIA’s core AI models to understand what evolution has already produced, rather than to create new entities. For models designed to generate molecules, the CEO says 99% or more of the chemistry they propose is constrained by the physics and energetics of making it; an AI-designed toxin, for example, would not necessarily be produced immediately.

The CEO also points to AI’s difficulty acting in the physical world: biological work can require custom workflows, and automating them remains difficult. A rogue human agent could cause greater harm, the CEO says, framing the risk as more human-centric than it is often portrayed.