Eric Schmidt and Suhas Mahesh argue that while narrow AI systems like AlphaFold have achieved breakthroughs, they depend on massive datasets that took decades to build and are impractical for most scientific fields. They contend that AI agents combining reasoning with tool access can model the iterative human research process, offering a more broadly applicable path to accelerating discovery. The authors say such agents could improve reproducibility, preserve institutional knowledge, and make hypothesis testing faster and cheaper across disciplines.
