MAIN CONTENT
Measured gains, not miraculous cures
The strongest case for AI comes from changes scientists can count. Novo Nordisk told Reuters that a design–make–test sequence once requiring roughly eight cycles can in some cases be reduced to three, while each cycle has shortened from about eight weeks to four. That does not guarantee a successful medicine: it means the company can explore possibilities more quickly and abandon weak candidates before investing further resources.
Novartis reported that the average interval from starting a project to identifying a drug candidate has fallen from approximately three years to between eighteen months and two years. Sanofi attributed at least a year saved in the development of the asthma treatment lunsekimig to an AI-assisted approach that helped researchers make earlier clinical decisions. These are company-reported examples, rather than independent proof of an industry-wide improvement.
Bristol Myers Squibb said its AI systems were instrumental in developing a sickle-cell disease candidate and are now being used to refine molecules and simulate choices with trial data. Researchers are also exploring how to connect predictive software to automated laboratories. The practical objective is not simply to generate more chemical designs, but to identify which ones deserve expensive physical testing.














