20VC: Mercor CEO on AI Moats, Agent Training, and Tokens Overtaking Salaries

The shifting power dynamics in the AI industry center on the diminishing defensibility of the software layer as models increasingly become the product itself. Brandon Foody, CEO of Mercor, argues that building lasting moats requires moving beyond raw data toward end-to-end systems that integrate expert human knowledge with automated quality checks and project management. While addressing recent security incidents and market rumors, Foody emphasizes that network effects and data moats will outperform application-layer companies that are vulnerable to model feature replication. A significant economic shift is underway as enterprises begin spending more on model tokens for internal agents than on human headcount, signaling a future where the primary job category involves training agents to handle repetitive execution. Mercor’s $10 billion valuation and network of 5 million experts position it to lead this transition toward high-value, multi-week automated projects in specialized fields like law and medicine.