Erik Torenberg

Erik Torenberg × Alex Rampell × Max Levchin

Visa and Mastercard still impose a roughly 2.5-second transaction window, a roughly 60-year-old standard that limits room for antifraud and payment innovation even as agents could reopen the interface.Affirm shows where the economics are strongest: Beautylish installments lifted conversion 30%, merchants fund negative CAC, and products up to 3½ years require machine-learning underwriting, while whether consumers will delegate purchase choices remains unresolved.

Erik Torenberg × Balaji Srinivasan

Srinivasan expects models to commoditize while trusted tribes retain personal, private, programmable context, shifting economic value from generation toward verification and expertise.AI may expand generalist agency and biology’s sensing potential, but physical bottlenecks, political resistance, distribution, and generic models limit the path to unconstrained autonomy.

Illia Polosukhin × Erik Torenberg × Nathan Labenz

NEAR’s user-owned AI thesis combines private, verifiable inference with decentralized compensation for GPUs, model weights, and proprietary data, with confidential computing offering 1% to 5% measured overhead.The network could turn benchmarks, fine-tuning, and even frontier models into inspectable, tokenized ventures, while unstoppable agents create unresolved risks around security, legal enforcement, human governance, and irreversible actions.