Core Theses & Mental Models
1. Market priors break: Discarding old venture maps
Old VC maps dictated that professional services (law, medicine), robotics, and company-built science were bad venture markets. Once foundation models cross a capability threshold, those historical “impossibles” become massive new market categories. Harvey, OpenEvidence, Sunday Robotics, and Chai Discovery are fundamentally the same bet in different industry costumes.
2. The untrainable corner: Where correctness is private
Thin wrappers get eaten by foundation models. Durable enterprise value sits where models cannot be scored from the outside: companies that own the proprietary definition of “correct” inside a customer’s private world—contracts, tickets, codebases, clinics, and factory floors. Sierra’s resolution-based pricing and Cognition’s software engineering guarantees are prime examples of capturing that last untrainable corner.