AI Agents, Not Just Data, Will Accelerate Science, Argues Eric Schmidt

In a Technology Review essay, former Google CEO Eric Schmidt and Suhas Mahesh argue that while DeepMind's AlphaFold demonstrated AI's potential in science, its data-intensive approach is not the best template for future breakthroughs. They contend that AI agents—reasoning engines with access to tools—will drive scientific acceleration by mimicking human reasoning under uncertainty. The authors cite AlphaFold's reliance on the Protein Data Bank, which took 53 years and $21 billion to assemble, as an example of the rarity of such conditions. They highlight Google's AI Co-Scientist, which independently formulated a correct hypothesis about antibiotic resistance spread that took Imperial College London researchers a decade to prove. Agents offer structural fixes for reproducibility, amplify scientific memory, and increase speed, potentially transforming science as profoundly as calculus or the computer. The essay was published on August 10, 2026.