“Every new medicine” is now designed on computers, the CEO of Bayer told Semafor World Economy on Wednesday, describing a fundamental shift due to AI.

“We used to have these banks of chemicals — thousands, tens of thousands of chemical entities — and we would screen them against targets,” Bill Anderson said, but now “we’re basically doing this with computational chemistry, computational biology.”

The same is true, he said, in crop protection. He said Bayer will launch “the first fundamentally new herbicide in about 40 years” next year, designed in silico: “It’s like the difference between shooting in the dark and having a precision-guided weapon.”

Since the unveiling of DeepMind’s AlphaFold in 2020, AI has vastly sped up the work of understanding suitable drug targets, especially by demystifying the structure of proteins.

DeepMind’s Demis Hassabis and his colleagues won the Nobel Prize for that work in 2024. But while it has removed one major bottleneck in drug discovery, the impacts in terms of actual new drugs have so far been relatively modest. Anderson’s statements suggest that is changing.

He noted that “A huge proportion of the interesting biological targets in medicine are so-called ‘undruggable’,” as in they are clearly important but it’s hard to get a drug molecule to attach to them. Now, a Bayer subsidiary called Vividion is using a “highly AI-dependent” method for developing drugs for previously undruggable biological molecules. “The first of these compounds are now in the clinic,” he said.