Drug discovery: the shift to agentic AI and self-driving labs

In Nature Chemical Biology, three researchers describe how agentic systems, coupled with benchtop automation, close the design-synthesize-test-analyze loop.

Pharmaceutical research is entering what three researchers describe as "agentic." In a commentary published on August 19, 2026 by Nature Chemical Biology, R. S. K. Vijayan and Jason B. Cross (Institute for Applied Cancer Science, MD Anderson Cancer Center) and Vasanthanathan Poongavanam (Uppsala University, SciLifeLab) argue that the convergence of AI agents and laboratory automation is shifting molecule discovery to self-driving labs.

The challenge is the DMTA cycle — design, synthesize, test, analyze — which structures the optimization of a drug candidate. Traditionally sequential and punctuated by human decisions at each stage, it becomes, with agents organized according to a ReAct-type architecture (alternating reasoning and action), a closed and adaptive loop: the system plans the next experiment based on the results of the previous one, without systematically going through an operator again.

The authors situate this evolution in the continuity of previous waves — QSAR models, deep learning, prediction of protein structures, virtual screening on a very large scale — but underline the difference in nature: it is no longer a model that predicts, it is a system that decides on the next experiment and executes it.