Frontier labs will split between intelligence-contingent domains and owning the human interface

pzakin · x · 2026-09-30

pzakin argues only a small set of domains (adversarial ones, recursive self-improvement) structurally reward frontier intelligence. Where commodity intelligence suffices, labs face strong downward pricing pressure, especially at the API layer. Application-layer ergonomics (habit, memory) shield against commoditization, and OpenAI is advantaged there since API revenue is a relatively smaller share. Long-term equilibrium: frontier labs compete in intelligence-contingent domains while fighting to be the principal human interface.

Original post →

More from Companies & People

Companies & People channel →