20VC: Mercor CPO on Revenue Concentration from Frontier Labs | Why Large Enterprise is Scared to Partner with Frontier Labs | Why Small Specialised Models is the Future with Osvald Nitski
Summary
Harry Stebbings interviews Oswald Nitzke, CPO of Mercor, on AI data strategy. They discuss whether open-source models cannibalize frontier demand, the shift toward specialized enterprise models, and ROI skepticism. Nitzke also covers evolving product management roles, hiring for judgment over tool usage, and the operational complexity of scaling human data services.
Topics discussed
Sponsorships and episode introduction
Open source models vs. frontier models and data value
Enterprise data sensitivity and model specialization
Navigating AI ROI and token spend economics
Mercor's hyper-growth and product surface area
Evolving product workflows and annotation platforms
Lab competition and the shift away from Figma
AI agent deployment: services vs. product
Hiring strategies and the Mercor Mafia
AI-fluent hiring and the importance of judgment
Product team structure and experimentation cadence
Scaling expert supply and retention strategies
Margins, revenue concentration, and self-serve goals
The complexity of human data projects and environments
Data pricing, unbundling, and competitive dynamics
Big Tech competition and the Threads precedent
Impact of the Mercor hack and cyber-defensive data
Talent wars, culture, and hiring 'assholes'
Quickfire round: advice, competitors, and mindset
Mercor's $200B case and the future of robotics
Listen ad-free on Castria