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

AI’s next battleground may be less about software features and more about who controls the models, the data, and the network. In this condensed recap of 20VC, Harry Stebbings speaks with Mercor CEO Brandon Foody about the company’s security incident response, why online rumors often distort the story, and how Mercor is building an end-to-end talent and AI project platform. They dig into the claim that application-layer companies have weak defensibility, the shift toward high-value multi-week work in software, finance, medicine, law, and consulting, and why “training agents” could become a major new job category. Foody also explains Mercor’s quality-driven moat, the likelihood of industry consolidation, and why infrastructure companies may outperform applications over the next year. Most strikingly, he says Mercor now spends more on tokens for internal agents than on employee headcount—a sign that AI economics are reshaping enterprise budgets. Listen now to get the key ideas in minutes.