Summary
Richard Socher, CEO of Recursive AI and former chief scientist at Salesforce, delivers a conference talk at AI Engineer about his vision for the “Eureka machine” — an AI system designed to automate scientific discovery across physics, chemistry, biology, medicine, neuroscience, and economics. The talk frames automated research as the next stage in a long arc of evolution-driven technological progress, connecting biological evolution, the Scientific Revolution, and the Industrial Revolution to today’s AI systems.
Socher outlines four pillars of the Eureka machine architecture: a foundation of codified existing scientific knowledge, integration of real-world measurement data, simulation capabilities for phenomena not yet directly measurable, and physical lab infrastructure for running experiments in the real world. Coordinating across all four layers is an agent swarm. He argues the bottleneck to scientific progress is no longer human intelligence but the fragmentation of knowledge across an ever-expanding number of niche subfields — a structural problem automated research agents are uniquely suited to address, as noted by Stanisław Lem decades ago.
Socher connects the work to you.com’s efforts building web search tailored for LLMs and agents, which he describes as meaningfully different from consumer search in structure and requirements. He mentions a forthcoming book on the Eureka machine concept. The talk is conceptual rather than a product demo, but offers a well-structured framework from an experienced NLP researcher and entrepreneur for where AI-assisted and eventually AI-led research is heading over the next decade.
📺 Source: AI Engineer · Published July 30, 2026
🏷️ Format: Keynote Launch
